N8123562
AN INTRODUCTIONTO THE INTERIM DIGITALPROCESSORAND THE CHARACTERISTICSOFASSOCIATED SEASAT SAR IMAGERY
SARTHE
JET PROPULSION LAB.
PASADENA, CA
01 APR 1981
https://ntrs.nasa.gov/search.jsp?R=19810015028 2020-05-06T15:58:38+00:00Z
JPL PUBLICATION 81-26
An Introduction to the Interim
Digital SAR Processor and theCharacteristics of theAssociated SEASAT SAR
Imagery
C. Wu
B. Barkan
B. Huneycutt
C. LeangS. Pang
April 1, 1981
National Aeronautics and
Space Administration
Jet Propulsion LaboratoryCalifornia Institute of TechnologyPasadena, California
ii'
/J
/
r ..
Tt_e research described in this publication was carried out by the Jet PropulsionLaboratory, California Institute of Technology, under contract with the NationalAeronautics and Space Administration.
TECHNICAL REPORT STANDARD TITLE PAGE
1. Report No. 2. Government Access;on No,._r'L Pub. _1-2_ ....
4. Title end SubtitleAn Introduction to the Interim Digital SAP.Processor and the Characteristics of the Associated
SEASAT SAR Imagery
7.C. Wu, B. Barkan/ B. Hune_cutt t C. Leang t S. Pang
9. Performing Organization Nc_.e and Address
3. Recipient's Catalog _.
5. Report DateApril 1, 1981
6. Performing Orqanizotion Code
8. Performing Orqanization i_eport 'No.
10. W'ork Unit No.
JET PROPULSION LABORATORY
California Institute of Technology4800 Oak Grove Drive
Pasadena, California 91103
"12"_ Sponsoring Agency Name and Address
NATIONAL AERONAUTICS AND SPACE ADMZNISTRATION
Washington, D.C. 20546
,5:Supp'em. ta .....
"11. Contract or Grant No.NAS 7--100
m
13. Type of Report and Period Covered
JPL Publication
14. Sponsoring _ency Code
RD1 5 l-j-_ 56.-_:-01-01-00,, _ ,,, , ,,
16. Abstract
An Interim Digital SAR Processor (IDP) was developed in 1979 to produce _ limited
amount of digital SEASAT SAR (Synthetic Aperture Radar) data to assist utility
studies in microwave remote sensing of earth resources and enviromnent. Thepurpose of this report is to provide basic engineering data regardin_ the Interim
Digital SAR Processor and the associated digitally correlated SEASAT SAR Imagery.Materials covered in this report include: I) SEASAT SAR processing funcSions,
2) An introduction to the Interim Digital SAR Processor, 3) IDP performance summary,and &) Characteristics of GEASAT SAR imagery. The first three subjects describe the
correlation function, IDP hardware/software configuration, a_i a preliminary per-
formance assessment. The last subject treats both the geometric and radiometric
characteristics of SAR imagery with emphasis on those that are peculiar to the IDPproduced SEASAT SAR imagery.
.q
17. Key Wor_ _elec_d by Authm_))Electronics and Electrical Engineering
Geosciences and Oceanography (General)Earth Resources
Computer Programming and Software
18. Distribution Statement
Unclassified - Un/_mited
19. Security Classif. (of this re_;rt) 20. Secur|ty C|mslf. (of this page) 21. No. of Pages 22. Price
Unclassified Unclassified• i i
li
_J
ACKNOWLEDGEMENT
The authors wish to thank R. G. Piereson for his thoughtful planning of this
development, J. C. Gilstrap for the design and implementation of the digital
tape-recorded computer interface, Dr. A. R. Johnston for the fiber-optic data
link, and Dr. A. E. Di Cenzo and Dr. D. N. Held for discussions and their
suggestion of an autofocus approach.
iii
ABSTRACT
An Interim Digital SARProcessor (IDP) wasdeveloped in 1979 to produce a
limited amount of digital SEASATSAR(Synthetic Aperture Radar) data to assist
utility studies in microwave remote sensing of earth resources and environ-
ment. The purpose of this report is to provide basic engineering data regard-
ing the Interim Digital SARProcessor and the associated digitally correlated
SEASATSARImagery. Materials covered in this report include: i) SEASATSAR
processing functions, 2) An introduction to the Interim Digital SARProcessor,
3) IDP performance summary,and 4) Characteristics of SEASATSARimagery. The
first three subjects describe the correlation function, IDP hardware/software
configuration, and a preliminary performance assessment. The last subject
treats both the geometric and radiometric characteristics of SARimagery with
emphasis on those that are peculiar to the IDP produced SEASATSARimagery.
iv
TABLE OF CONTENTS
I. INTRODUCTION .................................................... 1
2. SEASAT SAR PROCESSING FUNCTIONS ................................. 2
2.1 Introduction to the SAR Sensor Characteristics ............. 2
2.2 SEASAT-A Processing Functions .............................. 7
3. CHARACTERISTICS OF THE INTERIM DIGITAL SAR PROCESSOR (IDP) ..... 12
3.! Hardware Structure ........................................ 12
3.2 IDP Software Modules ...................................... 14
3.3 IDP Operation Procedure ................................... 26
3.4 IDP Output CCT Format ..................................... 27
4. IDP PERFORMANCE SUMMARY ........................................ 31
4.1 Resolution Performance .................................... 32
4.2 Sidelobe and Dynamic Range Performance .................... 32
4.3 Radiometric Calibration ................................... 34
4.4 Pixel Location Calibration ................................ 38
4.5 Throughput Performance .................................... 39
5. CHARACTERISTICS OF SAR IMAGERY ................................. 40
5.1 Geometric Characteristics ................................. 40
5.2 Radiome tr ic Characteristics ............................... 64
6. SUMMARY ........................................................ 80
REFERENCES .......................................................... 81
APPENDIXES
Ao
B.
C.
A Digital System to Produce Imagery from SAR Data ........ A-I
Statistical Characteristics of SAR Image Data ............ B-I
SEASAT SAR Radiometric Calibration Considerations ........ C-I
v
FIGURES
I.
,
3.
4.
5.
e
o
8.
9.
i0.
11.
12a.
12b.
13.
14.
A SEASAT SAR Ground Track (Rev. 1291) in the Northern
Hemisphere and SEASAT SAR Ground Receiving Stations ........ 4
SEASAT On-Orbit Configuration .............................. 5
SEASAT SAR Imaging Geometry ................................ 6
SEASAT SAR Doppler Characteristics ........................ II
Interim Digltal SAR Processor (IDP) Facility
Block Diagram ............... ............................... 13
Interim Digital SAR Processor (IDP) Software
Block Diagram ............................................. 15
Display of a Section of SEASAT SAR Raw Data ............... 17
A Simulated SEASAT SAR Point Target Response .............. 18
Single-Look and Four-Look SEASAT SAR Imagery .............. 22
Partitioning of Azimuth Spectrum for Multiple-Look
SAR Processing ............................................. 23
An Example of a I00 km x i00 km IDP Processed
SEASAT SAR Imagery ........................................ 25
2.6m Cubic Corner Reflector ............................... 33
Corner Reflector Array in a SEASAT SAR Image ............... 33
Measured IDP Gain Characteristics for Synthesized
Distributed Targets ....................................... 35
Measured IDP Gain Characteristics for Synthesized
Point Targets ............................................. 37
Geometrical Distortion in the Slant Range Imagery ......... 42
Slant Range and Ground Range Relationship for a
Spaceborne SAR ............................................ 44
vi
FIGURES(cont'd)
17. SARAntenna BeamConeAngle and Surface Incidence
Angle versus Slant Range for TwoSensor Positions ......... 45
18. SEASATSARGroundRangeResolution and Factor of
Conversion Expansion versus Slant Range................... 46
19. SEASATSARTarget Ground RangeDistance versus
Slant Range............................................... 48
20. Shortening Effect ......................................... 49
21. If the Surface Feature Slope is Larger than the
Radar Look Angle, Layover Occurs.......................... 51
22. Side Looking Imaging Radar Geometry Showing the
ShadowingEffect .......................................... 53
23. SEASATSARDoppler Center Frequency at Normal Attitude
versus Target Slant Range................................. 55
24. SEASATSARInstantaneous Doppler Frequency Response
Over Target Surface ....................................... 56
25. Location of SEASATSARFootprint as a Function of
Attitude .......................... _....................... 58
26. Geometric Distortion of IDP Processed SEASATSARImaging
and Distortion Due to Earth Rotation in a Line Scan
Imaging System............................................ 59
27. ImageData Block Skewing as a Result of RangeWalk
Compensation.............................................. 61
28. Speckle Reduction Through Image Smoothing................. 68
29. Corner Reflector Array and RangeSidelobes of 9m
Diameter Antenna.......................................... 72
vii
30.
31.
32.
33.
TABLES
i.
o
3A.
3B.
4.
SEASAT SAR Range and Azimuth Ambiguity Function for the
1645 Hz Pulse Repetition Frequency ........................ 74
Azimuth Ambiguity ......................................... 75
Azimuth Near Signal Suppression Effect .................... 77
Effect of Radar Aspect Angle to the Apparent Target
Reflectivity .............................................. 79
SEASAT SAR Sensor Characteristics and the Interim
Digital SAR Processor Performance Requirements ............. 3
List of Input Parameters for IDP SEASAT SAR
Processing ................................................ 24
SEASAT Image CCT Format ................................... 28
Reader Record Format ............... ;...................... 29
Summary of Image Parameters of Normal IDP Produced
SEASAT SAR Imagery ........................................ 31
viii
I. INTRODUCTION
The launch of the SEASATsatellite in June 1978 marked the advance of SAR
(Synthetic Aperture Radar) remote sensing technology from a conventional
airborne environment into an earth orbiting spaceborne environment. A
spaceborne SARcan provide wider width and much faster coverage capability
than can be provided by an airborne SAR. Thus a spaceborne SARoffers a
great potential for an operational global environmental monitoring system.
During the 105 days of SEASAToperation, approximately 50 hours of SARdata
were collected. The task of data reduction and utilization for SARtechnology
assessment remains an ongoing challenge.
While the bulk of the SEASATSARimagery was produced by an optical.
correlator, an Interim Digital SARProcessor (IDP) was developed to produce a
limited amount of digital SEASATSARimagery. The detailed design and
implementation phase began in April 1978. By March 1980, more than 150 SEASAT
SARframes had been digitally correlated and delivered to various users.
Current performance of the IDP meets the original design goals regarding its
image resolution capability and throughput rate. Weare also conducting an
IDP Upgrade Task to enhance the IDP throughput rate and image quality
performance.
f
The purpose of this report is toprovide basic technical information regarding
the IDP correlated SEASAT SAR imagery. Areas covered in this report
include: I) SEASAT SAR processing functions, 2) An introduction to the
Interim Digital SAR processor, 3) IDP performance summary, and 4) Character-
istics of SEASAT SAR imagery.F
2. SEASAT SAR PROCESSING FUNCTIONS
2.1 Introduction to the SAR Sensor Characteristics
A summary of the SEASAT SAR system parameters and the performance requirements
is given in Table 1 and reference [I]. Figure I illustrates a SEASAT orbit
track (Revolution Number 1291) and the associated SAR imaging swath (repre-
sented by the shaded area). Positions of SEASAT ground receiving stations and
their coverage are also plotted on the figure. Figure 2 shows the SEASAT
satellite configuration. One of the most prominent features is the 2m x 10m
SAR antenna which is mounted on the spacecraft with its boresight oriented at
a 20° angle from vertical direction. The received radar echoes were tele-
metered to the STDN ground receiver station over an analog data link. This
signal was digitized and recorded on a 120 M bits/s High Data Rate Recorder
(HDRR). The radar antenna was pointed to the right of the flight plath. The
SAR viewing geometry is illustrated in Figure 3. The antenna beam measured
approximately 6 degrees in elevation and 1 degree in azimuth. The radar
footprint on the Earth's surface (within the 3 dB beam contour) measured
approximately i00 km by 15 km.
A SAR nonnally utilizes an antenna pointing direction normal to the flight
path. Radar imaging in the cross-track (range) direction is accomplished from
the target echo delay, which is proportional to the distance from the sensor
to the target. Range resolution, which corresponds to the pulse width of a
point target response, is inversely proportional to the bandwidth of the radar
transmitted pulses. Radar scanning in the along-track (azimuth) direction is
accomplished by the satellite motion. The radar antenna beamwidth in azimuth
results in a footprint which is much wider than the range resolution
Table I. SEASATSARSensor Characteristics and the InterimDigital SARProcessor Performance Requirements
SARSystem Parameters (SEASAT-ASAR)
SAROrbit
Nominal Altitude
Nominal Speed
Transmitter Frequency
Pulse Repetition Frequency
Pulse Duration
Pulse Bandwidth
A/D Sampling Rate
(Range Offset Signal)
A/D Sampling Window Duration
Antenna Dimensions
Antenna Look Angle
Attitude (roll, pitch, yaw) Accuracy
Polar (108° inclination)
794 km
7450 m/sec
1275 MHz
1463, 1537, 1645 Hz
33.8 _sec
19 MHz
45.53 MHz
288 _ sec
2rex 10. Sin
20 ° elevation, 90 ° azimuth
+ 0.5 °
Interim Digital Processor Performance Requirements
Image Frame Size
Image Resolution
Number of Looks
Data Processing Speed
i00 km x I00 km
25 m
4
< 12 hours/frame
/60 /
/
170
®
/
/
Figure I. A SEASAT SAR Ground Track (Rev. 1291) in the
Northern Hemisphere and SEASAT SAR Ground
Receiving Stations
BUS
TRANET jBEACONANTENNA
TT&C _ANTENNANo. 2 SCATTEROMETER
ANTENNAS
SENSORMODULE
SYNTHETIC APERTURERADAR ANTENNA
-- _ TT&C
ANTENNA No. I _ /
VIRR RAOIOMETER
/SAR DA'_A - /
LINK ANTENNA
MULTI.CHANNELMICROWAVE RADIOMETER
LASER RETROREFLECTOR
ALTIMETER
Figure 2. SEASAT On-Orblt Configuration
_z
_0
\
\
\
Z
\\
\\
I.I
E!0
0l)
[--I
6
element. To achieve a comparable imaging resolution in azimuth, a correlation
process must be performed on the radar echo data to form a large synthetic
aperture along track, which produces a much narrower effective beamwidth.
Tutorial explanations on the SAR principles are given in References [2-4].
2.2 SEASAT-A Processin_ Functions
The major SEASAT-A SAR processing steps to produce high resolution imagery
include i) Range pulse compression, 2) Doppler parameter estimation,
3) Azimuth correlation, and 4) Multiple-look overlay. These functions must
incorporate several SEASAT SAR unique procedures to accommodate peculiarities
in the SEASAT SAR data (see para. 2.2.5).
2.2.1 Range Correlation
The purpose of the range correlation is to compress the time dispersed
(33.8 _sec) radar transmitted pule into an impulse. The 19 MHz bandwidth of
the transmitter pulse encoding enables a 6.6 meter resolution in the slant
range direction. This corresponds to approximately 17 meters to 23 meters
range resolution on the Earth's surface. The variation on the ground range
resolution is due to the different radar incidence angles over the swath.
2.2.2 Doppler Parameter Estimation
Synthetic aperture correlation is critically dependent on obtaining an accu-
rate estimate of the target phase delay history during the period of time that
a target is illuminated by the radar. The phase delay history of a target can
be accurately represented by a quadratic function with the instantaneous
Doppler frequency and Doppler frequency rate at the center of the antenna
illumination being two sufficient parameters to characterize this phase
history. Once these two parameters are obtained, a correlation reference
function can be derived. Therefore, the objective of the Doppler parameter
estimation function is to obtain accurate estimates of these two parameters.
2.2.3 •Azimuth Correlation
The azimuth correlation operation is performed to obtain a very narrow effec-
tive radar beamwidth in the azimuth dimension. To perform this operation, the
range correlated data lines are first accumulated to facilitate data access in
the azimuth direction. The result of azimuth correlation consists of four
independently correlated slngle-look images exhibiting approximately 25m
resolution in azimuth.
2.2.4 _itiple-Look Overlay
The multiple-look overlay function must accomplish accurate registration and
incoherent averaging of the intensity measures of the four single-look
images. The resultant intensity pixel values go through a square-root
operation and are linearly encoded into 8-bit data words.
2.2.5 SEASAT SAR Processing Peculiarities
SEASAT SAR processing differs from typical aircraft SAR processing in many
ways. Besides the unusually large pulse compression factors, three major
peculiarities are the severe range migration effect, the degree of uncertainty
in the SAR attitude, and the continuing change of target echo response func-
tion during each data collection pass.
In SEASATSARdata the range delay history of a point target traverses many
range resolution elements as the target passes through the radar antenna
beam. The migration history can be treated as a sumof two components: One
is directly proportional to the elapsed time or distance along the radar
flight path and is usually referred to as range walk. The other is propor-
tional to the square of the elapsed time and is called the range curvature
component. The parameters for these two componentsare generally referred to
as the Doppler frequency and frequency rate, respectively. For SEASATSAR,
the range curvature part of the delay history traverses approximately 8 range
samples from center to either edge of the radar beam. This implies that the
SEASATazimuth correlation must be a two dimensional process. The compu-
tational complexity is greatly increased by this amount of range curvature.
The uncertainty in radar attitude corresponds to residual error in the atti-
tude control angles. This error is approximately 0.5 degree in pitch and yaw
which is large compared to the I degree azimuth beamwidth. To achieve a good
signal to noise ratio in a SARimage, substantial data preprocessing is needed
to derive an accurate estimate of the instantaneous Doppler frequency at the
center of the beam. These estimates need to be periodically updated to
account for doppler variations.
The variation of the radar response parameters as a function of the sensor
orbit position is due to the fact that the surface speed due to earth rotation
changes with latitude. This results in a change of the sensor-target relative
speed and the associated Doppler frequency response. A plot of the center-of-
beamDoppler frequency and frequency rate as a function of the orbit anomaly
angle is shownin Figure 4. The SEASATSARprocessing parameters need to be
periodically updated to account for these parameter variations.
The data in Figur_ 4 also contain information regarding the magnitude and
bound of the range delay history. The functional relationship is given by
% 1 _At2), IAtl < 1.25 secR(At) = _ (fAt + _ (1)
where R(At) refers to the range delay relative to its center-of-beam value
(R(At = 0)), % is the radar wavelength, f is the center-of-beam Doppler fre-e
quency, and f is the Doppler frequency rate. The synthetic aperture integra-
tion time is approximately 2.5 seconds over the full aperture width• For a
value of f near 3,000 Hz, the total range migration can be 128 range samples
over the full aperture width. It follows directly from Eq. I that the syn-
thetic aperture phase history for the IDP azimuth correlation is defined by
Eq. 2:
1 _&t 2)_(At) = 2v(fAt + _ , IAt[ < 1.25 sec (2)
These equations show that accurate measures of f and f are sufficient to con-
trol range migration compensation and to generate the azimuth reference func-
tions within the SAR processor.
i0
o
I III
0o-c0
oo0
o00 E
uJ©
.gz
So_
o
Go
g
oco
o ioo
_?
o ¢_I1_ _l_
m
I'll _,t
0
0 0
11
3. CHARACTERISTICS OF THE INTERIM DIGITAL SAR PROCESSOR (IDP)
The Interim Digital SAR Processor (IDP) is a software based SAR processor
developed to produce digital SEASAT SAR imagery to support SAR application
studies. The throughput of the system is moderate, approximately one SEASAT
SAR frame (100 km x I00 km coverage, 25 m resolution, 4 looks) per ten hours
of processing time. The hardware component of the system is a general purpose
mini-computer based data processing facility. The software package was devel-
loped specifically for the IDP. A frequency domain fast correlation algorithm
was implemented to perform the SAR range _and azimuth correlation functions.
3.1 Hardware Structure
A block diagram of the IDP facility is shown in Figure 5.
sists of the following elements.
The hardware con-
3.1.1 Fiber-Optics Data Link
This fiber-optics data link provides the data transmission path between the
SEASAT-A HDDR (High Density Digital Recorder) and the IDP facility which is
located in a separate building at JPLo _
3.1.2 HDDR Data Interface
This interface hardware receives the bit-serial data stream from the HDDR
(through the optical data link), decodes the timing and interframe signals,
buffers each frame (one echo pulse), truncates the least significant bit of
each 5-bit data sample, and packs the resultant 4-bit data samples into 32-bit
words suitable for storage on the computer disk memory.
12
,,, t.) vu_l--
,"T 0"_
@
p.,.i.J-I-I,--I
P._
V
o_0
_JGo
,-4
.._ o_)
-,-I
_.l cJ._ o
u'3
13
J
3.1.3 SEL 32/55 Computer
This computer is a 32 bit minicomputer equipped with 96 thousand words of core
memory. It performs system level control functions for the software executionJ
and also performs a variety of data handling tasks.
3_1.4 AP-120B Floating-Point Array Processor
This array processor is equipped with parallel pipeline arithmetic adder and
multiplier and 16K words of high speed data memory. Its main function is to
perform vector arithmetic computations.
3.1.5 Disk Storage Devices
Two disk drives are involved°
is used for program storage and scratch-pad storage. The other disk has a
storage capacity of 300 M bytes. This disk is used for raw data storage
(220 M bytes space) and storage of intermediate computational results.
3.1.6 Other Computer Peripherals
Other IDP computer elements include a Dichomed image recorder and typicalp
Peripheral devices such as tape drives, line printer, and CRT control termi-
nals.
One has a storage capacity of 80 M bytes, which
3.2 IDP Software Modules
A frequency domain fast correlation algorithm was implemented in the IDP to
reduce SESAT SAR data into imagery. The basic processing algorithm concept
was originally reported in [5], which is also attached in Appendix A. An
updated algorithm description was reported in [6]. Major software modules and
their functional relationship are shown in a block diagram in Figure 6.
14
DA'I'APREPROCESSING
REFERENCEGENERATION
III-..II
I N PUTRAW DATA
RANGECORRELATION
CORNER TURNAND AZIMUTHTRANSFORM
AZIMUTHCORRELATION
MULTIPLE-LOOKOVERLAY
PIXELMERGE
100 km x 100 kmSEASAT-SARIMAGE FRAME
" IMPULSE-RESPONSEGENERATOR
DATAVERIFICATION
• RAW DATA IMAGE
o HISTOGRAM
• LINE AVERAGE
• POWER SPECTRUM
Figure 6. Interim Digital SAR Processor (IDP) Software
Block Diagram
. 15
f
A brief summary of the function of each program module is provided in the
following subsections.
3.2.1 SAR Raw Data Input Program
This program accepts the packed 4 bit raw data samples and loads them onto the
300M byte storage disk. A photographic representation of a_typical raw data
set is shown in Figure 7.f
3.2.2 Point Target Response Simulation Program
This program simulates the point target echo response of the SEASAT-A SAR
sensor. The program was used to test and verify the SAR fast correlation
programs. A synthesis of a typical SEASAT SAR point-target response is shownf
in Figure 8.
3.2.3 Data Test and Verification Prosram
This program comprises a number of subprograms which include the data/image
display program, the histogram program, the power spectrum program, the data
averaging program, etc. This program is used to verify the data quality at
each stage of SAR correlation processing.
3.2,4 Data Preprocessin$ Program
This preprocessing program is used to estimate the key parameters for the
azimuth correlation of the SEASAT-A SAR data. Parameters to be estimated are
f
the instantaneous Doppler frequency and Doppler frequency rate of targets near
the center of the antenna illumination. Estimation of Doppler center fre-
quency can be made on the radar raw data by first obtaining the power spectrum
16
Figure 7. Display of a Section (2048 x 2048 Samples) ofSEASATSARRawData
17 _ Reproduced from _._b'est available copy_ _._
Figure 8. A Simulated SEASAT SAR Point Target Response
Pattern (2048 x 2048 Samples, Approximately One-Half
of SEASAT SAR Response)
18
along the azimuth dimension. A more accurate method is used here which
analyzes the intensity of corresponding pixels over the four azimuth looks.
An error in the Doppler frequency rate affects the synthetic aperture focusing
and therefore the registration accuracy of the four-look processing. The
auto-focusing method adapted in the IDP to refine the Doppler frequency rate
estimates is based on the auto-correlation of segments of one-look pictures
and cross-correlation between the four slngle-look images of the same area.
3.2.5 Reference Function Generation Program
This program generates the reference functions for azimuth correlation, based
on the estimated Doppler parameters provided from the data preprocessing
program.
3.2.6 Range Correlation Program
This program performs the SEASAT-A SAR data range offset to I, Q (inphase and
quadrature) conversion, the range correlation, preliminary range walk correc-
tion, and the first-stage corner-turn functions. The fast correlation is
performed on a block of 4096 4-bit real raw data samples. Range offset to
I, Q conversion is performed in the frequency domain by a translation of the
offset spectrum to baseband. The range reference function has a length of
768-complex samples. The result of the correlation for each 4096 point input
data block is 1280 complex data points. The real and imaginary parts of each
complex word are truncated and packed into a 16-bit word, i.e. 8-bit for real
and 8-bit for imaginary. During the course of range correlation the starting
sample of the 4096 real data block is systematically adjusted to achieve first
order range walk correction. The range reference function is also adjusted to
provide a finer step size in the range walk correction. A total of four range
19
reference functions over a real range offset data bin are used to provide a
range walk interpolation to I/8 of the size of a complex range bin. The range
correlated data are first stored in a 32-1ine buffer storage. They are
regrouped into 40 each 32 x 32 point blocks and then output to the disk stor-
age. This blocking facilitates the subsequent corner turn operation.
3.2.7. Corner Turn and Azimuth Forward Transform Prosram
This program calls se%ected blocks of data as described above, stores them in
a buffer, and accesses the data in the azimuth dimension. This accomplishes
the data corner turn operation. A Fourier transform then is performed on
these azimuth data vectors° The resultant spectra, which are packed into
16 bits per complex word, are output to another disk storage file. The
dimension of the data vector to perform the Fourier transform is 2048 complex
samples in azimuth. _i
3.2.8 _zimuth Correlation Program
This program calls the azimuth spectral data, performs a range migration
compensation in the azimuth spectral domain, multiplies the spectral data by
the azimuth reference transfer_function, and then applies an inverse Fourier
transform to the product. These filtered complex data go through a power
detection process. The results correspond to a single-look imagery. The
range migration compensated spectrum is a vector containing 2048 complex
terms. The length of the azimuth reference function for each look spans 1024
complex samples in azimuth. Because each single-look image is produced by
processing one quarter of the total synthetic aperture (or one quarter of the
azimuth spectrum as is defined by the antenna radiation pattern), the inverse
Fourier transform of the fast correlation is performed on a 512 term vector.
20
256 samples are retained after the inverse transform, and are detected to form
256 one-look pixels. A total of four 512-vector inverse transforms are
performed to obtain four sets of single-look pixels. The correlation vector
size (2048) and the reference function length (1024) are such that these four
single-look line segments (256 pixels each) register side by side to form a
1024 point one-look image llne. An example of a single look image (1024 lines
and 1024 azimuth picture samples per line) is shownin Figure 9. Each of the
four vertical sections is of one particular single look. The effect of
subdivision of the azimuth spectrum, which results in a stronger response in
the center two looks and a weaker response in the two side looks, is clearly
evident in this picture. A graphical illustration of the four single look
spectral bands over the azimuth spectrum is shownin Figure I0.
3.2.9 Multi-Look Overlay Program
This program performs the four look overlay of the single-look images produced
by the azimuth correlation program. The pixel intensity values are averaged
to produce the four-look imagery. A square root operation is applied to the
intensity value of the final four-look pixel. The resultant value is encoded
linearly by a 8-bit pixel word. The 8-hit amplitude representation is capable
of providing an output dynamic range of approximately 48 dB.
3.2.10 Pixel Me[ge Prosra m
This program merges the multiple-look overlayed image blocks, each of which is
256 (azimuth) by apprdximately Ii00 pixels, into an image of approximately
5800 lines and 6144 pixels per llne. The resultant image corresponds to a
SEASAT-A SAR frame covering a i00 km x I00 km area on the ground surface, and
21
(a)
(b)
Figure 9. Single-Look and Four-Look SEASAT
SAR Imagery (1024 x 1024 Pixels)
(a) Single-Look Image
(b) Four-Look Image
22
\
AZIMUTHSPECTRAL
REPETITWERESPONSE
SPECTRAL PATTERN _DUE TO DISCRETE /
PRF 5A_
_%_I OOK I LOOK 2
PRFC
DOPPLER CENTER
FREQUENCY
PROCESSING BANDWIDTH
(APPROXIMATELY 1300 Hz)
,:,.i
Figure i0. Partitioning of Azimuth Spectrum for
Multiple-Look SAR Processing
23
has 25m resolution, 4 looks.
SEASAT SAR imagery is shown in Figure Ii.
California Coast, south of Los Angeles.
25, 1978, SEASAT revolution number 1291.
An example of such a merged I00 km x I00 km
This image covers a section of the
The data were obtained on September
3.2.11 Executive Program
This program serves as the executive controller for the sequential execution
of the correlation processing programs defined in 3.2.5 to 3.2.10. The set of
parameters entered into this executive program to initiate the IDP processing
are listed and explained in Table 2.
Table 2. List of Input Parameters for IDP SEASAT SAR Processing
LBIAS
NBIAS
ISTEP*
RINC*
RCHG •
LAZI
FD
FDD*
FRATE
FRCON
PD
Starting Line Number of raw data stored on disk
Starting sample number of each record
Amount of range walk compensation over 1024 lines
Sign of sample increment for range walk compensation
Period for increment (= 1024/ISTEP)
Number of azimuth lines to process in each look
Doppler center frequency
Center frequency offset for range delay calculation
associated with range walk compensation
Doppler frequency rate over synthetic aperture at near swath
Decrement of Doppler frequency rate every 64 pixels in range
Reference weighting coefficient for sidelobe suppression
"r
*Parameter whose value dependent on ISTEP and can be eliminated.
24
Figure ii. An Exampleof a i00 km x i00 km IDPProcessed SEASATSARImagery
25
3.3 IDP Operation Procedure
To produce a I00 km by 100 km SEASAT-A SAR imagery on the IDP, the standard
procedure comprises the following steps:
I) Obtain the earth coordinates, the SEASAT SAR experiment site, and
any additional constraints regarding the time or conditions of the
radar observation. Select a SEASAT orbit revolution containing
data from the experiment site.
2) Obtain the starting time of that revolution from the Catalog of the
SEASAT SAR Experiment Sites [7]. Next, calculate the time at which
_the satellite began to collect data from the experiment site.
3) Play the HDRR tape corresponding to the selected orbit revolution
and execute the SAR raw data input program (3.2.1) to store a
prescribed set of SEASAT-A radar data on the 300 M byte computer
disk memory. The data begin at the time determined in Step 2 and
include an amount that corresponds to a data acquisition period of
approximately 18 seconds.
4) Execute the data preprocessing program (3.2.4) to obtain accurate
estimates of SEASAT-A SAR azimuth correlation parameters.
26
5) Enter the parameters from Step 4 into the IDP executive program and
initiate its execution. The current system incorporates 15 loops
of the correlation programs -- 3.2.6 to 3.2.10. The result of each
loop is a block of four-look imagery of approximately 1200 by 2048
pixels. These image data are output to an intermediate computer
tape.
6) Initiate a second state of the merge program which merges the 15
sub-blocks of the processed ii_gery into the final image format
which has approximately 5800 azimuth records and 6144 samples per
record. This merged I00 km x I00 km SEASAT-A SAR image frame is
stored on a 1600 BPI (bits per inch), 9-track computer compatible
tape (CCT).
3.4 IDP Output CCT Format
The format of the IDP output tape which contains a 100 km x i00 km SEASAT-A
SAR image frame is provided in Table 3.
27
Table 3A. Seasat ImageCCTFormat
(File i)
HEADER
DATALINE 1
DATALINE 2
DATALINE 3
Record
Record
Record
LASTDATALINE Record
TM
(File 2)
HEADER Record
DATALINE 1 Record
TM
(Last File on Tape)HEADER Record
DATALINE 1 Record
TM
TM
28
t",'h
U.1,.2
,<
m
s
ot.._
0
I
4=I
O
t._
0
o_
l.d
JJI.,...I
I
:> Q;
_=_ _J
0
0 _=_ '_ _
I) il _ U
u_ _0 _I_I I I
u_ _ u'h
C_ I: 0
_ ,,-.__ .=_ 0•"_ 0 • _ ,._
II II II II 11
_..-_I._
.._
m_._
0 W
I_ I,.,
I.,,_=
I
,.O
I
oO
<:
r.,o
°r.,o .. _J
_ s
r_ ._ _ o
_ ,,=1
IZ
_ m
¢; _ 0 0
P_ I1 II
I
0
Z
0,1I,.u
0
I",,.,
29
I
I
E
t-o
.,-d
4.J
(p
II II
X
x _
E E
,_ oo E E
•,.1- _, _ _._
oo
I
0",
°,,-I
O
e_
m,
T , 7 7 - - ? _I I I
o
(=
..I=I
o_==_
I=,
E
.I.=I
i.Ir.I
0
F_
C
o,=_
_0
0
OJ
_J
a.J _ O;
o _- o
O)
•,_ _ 0 ",_ .,* _) • ,-_
O0 O_ _ ,-, C,4 _ "4" u'_ %ID
30
4. IDP PERFORMANCE SUMMARY
The nominal performance of the IDP system in producing SEASAT SAR imagery is
summarized in Table 4. Several key issues in the performance evaluation are
discussed here.
Table 4. Summary of Image Parameters of Nominal IDP
Produced SEASAT SAR Imagery
Input Raw Data
Range Resolution
Azimuth Resolution
Range Peak Sidelobe ratio
Azimuth Peak Sidelobe ratio
Number of Looks
Pixel Dynamic Range
Pixel Radiometric Accuracy
Pixel Geometric Accuracy
4 bits/sample
25 m
Approximately 25 m
-15 dB
-6 to -9 dB
4
Selectable 48 dB (over 70 dB
total) in 8 hits amplitude
1TBD
TBD 2
I. Processor gain over swath depends slightly on the input Doppler center
frequency value of its closeness to the Doppler centroid. Uncertainty is
mainly associated with gains in radar sensor and SAR data communicationlink.
2. Knowledge of pixel relative position is likely within + I00 m over a flat
area. Absolute accuracy can be much poorer due to uncertainties in the
platform position and data timing which remains to be calibrated.
31
4.1 Resolution Performance
The IDP processing algorithm as described in Appendix A is capable of achiev-
ing a SEASAT SAR image resolution of 25 meters in both range and azimuth. A
routinely produced image may have a resolution slightly worse than 25m due to
error in estimating SAR processing (focusing) parameters. The IDP processed
corner reflector array scene near Goldstone Tracking Station is shown in
Figure 12a. The corner reflectors, each of which has a dimension of 3m or
less, appear as distinctive points in the image. Interpolation of plxels near
the peak responses of the reflectors indicated a 3 dB resolution of approxi-
mately 25m in range and 35m in azimuth. A refined preprocessing procedure
installed in June 1980 yields consistently highly accurate estimates of pro-
cessing parameters. Resolution broadening due to residue error in the
estimated parameters obtained by the refined process is expected to be within
5m.
4.2 Sidelobe and Dynamic Range Performance
Detailed performance evaluation on intergrated sidelobes, dynamic range,
contrast ratio, and image modulation transfer function is being conducted at
the present time. By using a step-function to approximate the range curvature
as described in Appendix A, the integrated side-lobe ratio is of approximately
-6 to -9 dB. For truely distributed targets, e.g. ocean, this integrated-
side-lobe ratio implies an image contrast of approximately 6 to 9 dB. The IDP
produced plxels are represented in amplitude format. The contrast or side-
lobe ratio observed in the amplitude domain is half of the dB values discussed
above. The maximum possible dynamic range for an eight-bit pixel in amplitude
representation is of 48 dB which is derived from the ratio of the biggest
value 255 and the minimum discriminatory interval, i.e. I.
32
Figure 12a. Picture of a 2.6m Cubic Corner Reflector
Figure 12b. Corner Reflector Array in a SEASAT SAR Image
33
4.3 Radiometric Calibration
SAR data input to the IDP are quantized to a 4_ bit integer with a range frequency
offset as described in the previous section. Each four bit integer assumes a
value from 0 to 15. These positive integers can also be treated as bipolar
signals of mean value 7.5, and a data value ranging between + 7.5. The theo-
retical gain of the SEASAT SAR processing is the product of the following three
factors: time-bandwidth product of the range linear FM-chirp waveform, time-
bandwidth product of the azimuth chirp waveform per look, and the square root
of the number of looks. The numerical values of these three factors are of
approximately 642, 200, and 2, respectively. The intensity gain of a pixel
response from its strength in SAR echo signal to that on the processed imagery
is therefore 54 dB. The corresponding gain in amplitude strength is 27 dB.
The amplitude gain of IDP is calibrated using synthesized SAR echo signals of
distributed and point targets. For distributed targets, Gaussian random noise
of the same mean value of the SAR raw data but various standard deviations
were generated and quantized into a 4-bit integer format compatible to the IDP
input. Saturation was applied to data values outside of the limits of 0 or
15. These synthesized data were processed through the IDP. The square root
of the mean intensity of output pixels (mean of the squared 4-1ook pixel
values) produced from the synthesized input versus the standard deviation of
the unsaturated noise is plotted in Figure 13. The SAR processor performs
linearly to a level of approximately 17 dB (square root of the mean pixel
intensity). Saturation of the mean output level is definitely due to the
limited range of the 4-bit quantized input data. The difference between the
output level and the standard deviation of the input over the linear range is
the calibrated amplification of the IDP system, which is observed to be
34
0-
15-
MEAN OUTPUT
LEVEL
17.2 dB LEVEL OF SATURATION
[5
(dB)
I I10 15
INPUT NOISELEVE L" a"
I
2O
Figure 13. Measured IDP Gain Characteristics for
Synthesized Distributed Targets
35
approximately I0 dB. For distributed targets, this measured I0 dB gain
implies that the meanecho response (in quantized input domain) is
approximately I0 dB below the observed square root of the meanpixel
intensity. The expected gain is 12 dB due to a 16-fold controlled increase in
data value in range correlation. The difference could mostly be attributed to
the finite processing bandwidth in both range and azimuth, and data saturation
due to limited number of quantization bits.
The quantized echo data are superposition of returns from targets over the
entlre radar footprint. The SARecho response of a single point target of
amplitude muchweaker than the minimumquantization level I (over the input
range 0 to 15) can not be synthesized (due to this initial raw data quanti-
zation) without readjustment of the IDP processing gain. The synthesized
point target responses of various magnitudes were generated and superimposed
by noise (4 i 2). The resultant pixel values (square root of the four-look
summedintensity) versus the original point target amplitudes are plotted in
Figure 14. The two dashed lines in the figure are the expected background
level due to the finite additive noise, and the maximumoutput level which
corresponds to 255 in amplitude value respectively. The upper line at a 45°
angle to the axis indicates the expected gain performance 27 dB + I0 dB for
point targets exhibit specular reflection. The dotted 45° line indicates the
expected gain of a speckly target, which is 1.5 dB below the ideal point
target gain due to expected speckle noise variation. The measured points
suggest near linear gain performance below the 8-bit saturation level. A 2 dB
loss in the measuredtarget response in the linear region relative to the
expected gain performance (solid line) is unclear and is being analyzed.
36
i
POINT TARGET_
OUTPUT LEVEL T_
EXPECTED PERFORMANCE OF _/"
SPECULAR POINT TARGETS ,_/ r _
-30 ' -25 -20 -15 -10
(dB)
3O
24 dB (=255) PEAKSATURATION LEVEL
--_v_- ----_ _ -X---
-15 13 dB MEAN NOISE LEVEL(INPUT NOISE _r = 3 dB)
I0
INPUT AMPLITUDE OFPOINT TARGETS
I , i c_5 10
Figure 14. Measured IDP Gain Characteristics for Synthesized
Point Targets ( e: fD = 0 Hz, x: fD = i000 Hz)
37
Possible sources of this difference include: course quantization in input
which affects the meaninput and the detected energy, and interference effectp_
due to noise. This figure nevertheless indicates the relationship between an
individual pixel value and the expected echo signal strength of that pixel in
the SAR raw data.
The SAR processor is one element of the entire SAR system. Accurate
radiometric calibration of the resultant imagery requires knowledge of the
gain function of the radar transmitter/receiver sensor and the SAR
communication link system. Discussions on such gain functions made by
B. Huneycutt are included in Appendix C.
Radiometric accuracy is also affected by the resolution performance of the
IDP. The effect is significant for point targets because a degraded reso-
lution also results in a loss of target intensity response. The loss of gain
of a point target due to a degraded resolution corresponds roughly to the
ratio of broadening to the ideal resolution.
4.4 Pixel Location Calibration
The absolute location determination of a&l pixels based on SAR processing
parameters and orbit determination data on the Sensor Data Record (SDR) is
observed to contain + 2 km uncertainty. The current source of error is
speculated to be in interpretating the satellite ephemeris data. A study is
being conducted tO resolve this error. The pixel location procedure for
absolute location determination can be used for each individual pixel on the
image. Using this absolute location procedure, the relative pixel locations
38
within a I00 km x I00 km SEASATSARframe can be determined to within + 25 m
accuracy for targets located in a very flat area.
4.5 Throushput Performance
The current IDP processing time to produce each i00 km x I00 km, 25 m
resolution 4 look SEASAT SAR image frame is approximately I0 hours. The
processing is done over night shift hours. The average throughput rate is
approximately 3.5 SEASAT SAR frames per week.
39
5. CHARACTERISTICSOF SARIMAGERY
Radar imagery superficially resembles conventional aerial photography but
differs from it in many respects. This section provides a brief summaryof
the basic characteristics of SARimagery. The emphasis is on those parti-
cularly related to SEASATSARimage interpretation. The geometric and radio-
metric aspects of SARimage characteristics will be treated separately in the
following discussion.
5.1 Geometric Characteristics
SAR imagery is generally represented in a two dimensional format representing
the along-track and cross-track directions. Geometric distortion refers to
change of location of a target as it appears in a SAR image with respect to
its location in a scaled geographic map. The most predominant factor which
causes the geometric distortion in a typical SAR image is that the cross-track
distance to a target in radar imagery usually corresponds to the slant-range
distance between the target and the sensor. Because the radar sensor is
placed at a position much above the ground surface, the slant-range distance
measurements to individual targets do not relate linearly to the corresponding
distance measurements on the planet surface. Effects due to such range dis-
tance mapping may include: ground range nonlinearity, range foreshortening,
radar layover, and radar shadow. These effects are discussed respectively in
subsections 5.1.1 to 5.1.4. SEASAT-A SAR imagery also exhibits a geometric
distortion due to earth rotation. This effect will be discussed in section
5.1.5.
40
5.1.1 Slant Ranse/Ground Ranse Nonlinearity
Figure 15 illustrates the difference between a ground range (distance measured
on the ground surface) and a slant range (distance measured from the sensor to
a target) image. A and B, represent two features of equal size under radar
illumination. Radar raw data samples are typically equally spaced in slant
range distance. In this slant range display of radar imagery, those two
features, A and B, will be compressed with a factor which depends on the posi-
tion of the feature in the beam. Objectives in the near range are more com-
pressed than those in the far range. The difference in appearance of a square
grid containing linear and circular features in these two representations is
also shown in Figure 15.
For spaceborne SAR imaging over a curved (spherical) earth surface, the
relationship between a small increment in surface range, dx, to its
corresponding increment in slant range, dr, (see Figure 16) can be expressed
as follows:*
(H + R )sin0
e dx (3)dr = dx sin@ = R
e
with
2 2 _ R 2 ..
-1 r + (H + Re ) e (4)e -- cos 2r(H + R )
e
where _ is the radar angle of incidence, @ is the elevation look angle, H is
the radar altitude, R is the radius of earth, and r is the slant range.e
*In the following analysis, second order effects due to earth rotation and
atmospheric refraction are neglected.
41
S PAC[C RAFT
; GROUND RANG[ I_,G[ __i
/
//
GRO(/ND RANGEIMAGE
"-. NEAR
- -.@ANGE
I/l
)\ /
SLANT RANGE
FMAG[
Figure 15. Geometrical Distortion in the Slant Range Imagery. Near Range
Features are Compressed Relative to Far Range Features. Correc-+
tions During Ground Data Processing can be Applied to Generate
Ground Range Images
42
Since the local value of R and the value of H can be determined from thee
satellite orbit data, and the slant range distance can be accurately deter-
mined from the delay timing of the echo, the value of dr/dx is readily deter-
mined. Ths factor also indicates the relationship of radar resolution in
slant range to the resolution in ground range. The slant range resolution,
_r , is a constant as derived by the 19 MHz SEASAT pulse bandwidth, i.e.
C
Ar -- 0.88 2--B = 6.95 m (5)
where C is the speed of light. The ground range distance X as a function of
the slant range r (see Figure 16) can be expressed as:
r
X = R 6 = R sin-l(_ -- sin@) (6)e e
e
where angle 6 is the one spanned by ground range distance to the center of the
earth, and @ is the beam elevation look angle defined in Eq. 4. The following
inverse relationship can be derived by trigonometric means:
i/22 x__}= {Re + (Re + H)2 - 2Re(Re + H) cosr
Re(7)
Eq. 7 is particularly useful in converting slant range representation to
ground range.
Assuming a nominal SEASAT altitude of 794 km, and a radial distance from the
surface to the earth's center of 6369 km near 40 ° latitude, the beam incidence
and elevation angles versus slant range distance are plotted in Figure 17.
Figure 18 contains curves which describe the relationship between slant range
resolution and ground range resolution for selected cases. Figure 16 contains
43
SENSOI
_h.\ dr /
NADIR iI__/'_,_//
Re
0
Figure 16. Slant Range and Ground Range Relationship
for a Spaceborne SAR
44
29--
28--
27--
26--
25-
24-
23-
22-
21-
20-
19-
18-
17--
16--
15130
----- ANTENNALOOK ANGLE
-- ANGLE OFINCIDENCE
//
//
/
/
H = ALTITUDE
= ORBIT ANOMALY
(ORBIT INCLINATION = I08 °)
I I I _ I850 860 870 880 89084O
SLANT RANGE R0 (kin)
Figure 17. SAR Antenna Beam Cone Angle and Surface
Incidence Angle versus Slant Range for Two
Sensor Positions
45
EZO
..J
O
uJ
L,J
OZ
Z
O
O
28
26-l
24-
22-
20-
18-
16-
1483O
)
H = 789 km / ¢_= 90°)__
H = ALTITUDE
at = ORBIT ANGLE(ANOMALY)
1 I I I 1840 850 860 870 880
SLANT RANGE R0 (kin)
4.0
3.5
3.0
2.5
Z<Iv"
Z
O(.0
Ok--
Z
n,,"
O
ZO¢,/3
Z<
Xl.,I.I
L4.
Om,,"
O
u
Figure 18. SEASAT SAR Ground Range Resolution and Factor
of Conversion Expansion versus Slant Range
(Pixel Spacing on Surface Equals 6.6m Times
the Conversion Expansion Factor)
46
curves which describe the relationship between the slant range and the
distance from the nadir point for selected cases. These three figures provide
some information regarding the geometric aspects of SEASAT SAR imagery in
slant range representation. The basic pixel spacing in slant range is a
constant of 6.59 meters. Figure 18 indicates that range pixel spacing on the
surface varies approximately from 20m to 15m from near swath to far swath.
The ground range distance versus slant range can be determined from an
integral of the rate of expansion function described in Figure 18. The slant
range to ground range relationship exhibits a small amount of nonlinearity as
shown in Figure 19.
Conversion of slant range format to ground range can be achieved through a
resampling process. Calculation of the positions of new samples in slant
range that correspond to equally spaced data samples on ground rac_ge _akes use
of the formulas given in Eq. 7. Methods of resampling were suggested in
Ref. [8].
5.1.2 Radar Foreshortenin$ !91
Radar foreshortening, a distortion inherent to all radar imaging of
irregularly shaped terrain surfaces, is the variation in the apparent length
of equal terrain slopes when the slope measurements are taken at different
incident angles.
Figure_ 20 illustrates the time intervals between equally spaced radar energy
pulses which intercept the terrain at a certainrange from the spacecraft. On
radar imagery, the distances between terrain features are recorded as a func-
tion of the time interval required for the RF energy to tranverse the
4 7
\
I I B I I I
-- oo
o
o
E
0
-_e_
7.
-_,
0o3C
t_C E
c
m
CC
"_ C o
_ m_'_ _ o
m t_ __ o
< N
48
/ /,
x/" /" x,_;, i"/ _,/ /" _ -/ /
/.////// /// //'_.//_/..
/ / .,,_"//_/- / Y/ / .tf//" "-,'7'/'_ / /
Figure 20. Shortening Effect. Slopes Inclined Toward the Radar Look
Shorter Than the Slopes Inclined Away From the Radar
49
corresponding separation distance. The actual lengths of terrain slopes ab
and bc are equal; however, in the radar imagery, the length of slope ab would
appear considerably shorter than bc (a'b' < b'c'). The dimension
transformation is given by
\
a'b' = ab sin(@ - _ ) (8a)s
b'c' = bc sin(_ @ 8 ) (8b)s
thus, terrain surfaces sloping toward the radar will appear shortened relative
to those sloping away from the radar.
The radar foreshortening effect is minimal at large incidence angles
--4
(_ = 90 ) and maximal at small incidence angles (_ _ 0 ) • Before reaching
these limits, however, large radar layover and shadowing effects appear.
5.1.3 Radar Layover [9]
Radar layover is caused by the radar signal intercepting the top of a feature
before it intercepts the bottom or, more precisely, when the terrain slope is
greater than the incidence angle _. As a result, the objects imaged appear to
"lean" toward the nadir of the spacecraft. This effect is more likely to
occur when imaging terrain features with appreciable relief (such as mountain-
f ous areas), and especially if these features occur in the near range. I
As illustrated in Figure 21 the tops of terrain features i and 2 are recorded
on the imagery before the bases because the slant range distance from the
radar to the top of the features R t is less than the slant range distance R b
50
SPACECRAFI
HI
/
Figure 21. If the Surface Feature Slope is Larger than the Radar Look Angle,
Layover Occurs. In This Case the Signal Scattered From the Top of
the Feature is Received Before the Signal Received From its Bottom
(Rt_ _).
51
to the bottom. With terrain feature 3, the slant range distance Rt and Rb are
approximately equal and therefore the slope is imaged nearly as a single
point, or a line if the feature is elongated in the along-track direction.
Terrain feature 4 is imaged normally; however, radar foreshortening now
affects the apparent length of terrain slopes as discussed previously.
5.1.4 Radar Shadow [9]
Shadows which are usually present in aerial photographs are a function of the
position of both camera and sun. Since radar provides its own "illumination,"
radar shadows appear at the side of a terrain feature most distant from the
spacecraft whenever the terrain back slope exceeds the complement of the
incidence. Radar images of terrain made at comparatively large look angles
form shadows on the imagery that are analogous to those shadows formed on
aerial photographs taken at low sun angle. Figure 22 illustrates the shadow-
ing effect.
5.1.5 Geometric Distortion Due to Earth Rotation
This subsection treats a geometric distortion caused by various factors which
affected the Doppler shift associated with individual targets as they passed
through the SEASAT antenna beam. Two main factors in the distortion are earth
rotation effects and the fact that the effective beamwidth of the synthetic
aperture does not align with the satellite cross-track direction. The
rotation effect is common to satellite scanning systems where the scan lines
are gathered serially in time while the earth surface moves continuously
during the imaging period. The location and orientation of the SAR image
lines are strongly affected by the Doppler parameter value used in the syn-
thetic aperture processing. The steering of the synthetic beam is subject to
52
SPACECRAFT "
I
two practical constraints: One is that the synthetic beam must be very close
to the-center of the physical antenna beam in order to attain an adequate
signal-to-noise ratio in the imagery. However, the footprint of the physica !
antenna beam is often not aligned with the cross-track direction due to an
error in the antenna pointing relative to the nominal azimuth angle of 90 °.
The other constraint pertains to the complex Doppler response_ of targets in
the antenna footprint. Because of the effect of earth rotation in conjunction
with the curvature of the earth's surface, the Doppler frequency response
depends not only on the target latitude but also on the location of the target
in the I00 km SEASAT SAR swath. An illustration of the Doppler frequency
response as a function of the orbit angle and the spacecraft attitude was
given in Figure 4. The Doppler frequency_characteristics as a function of
target swath position and latitude (assuming perfect antenna pointing) are
plotted in Figure 23. By adjusting the SAR processing to reflect the Doppler
frequency change over the swath width, it is possible to align the SAR image
lines to the center Of the physical antenna footprint. However, it is
extremely difficult to implement a digital SAR @rocessor which adapts to this
change of target center frequency over a swath without introducing undesirable
artifacts or excessive complexity. The accuracy of multilook registration or
the image resolution can be adversely affected. Further, without knowing the
motion of the sensor to a high degree of accuracy, there always remains a
residual error.
An example illustrating the dependence of target Dopper frequency on the
target location (relative to the satellite track) is described in Figure 24.
A plot of the SAR antenna footprint locations corresponding to several differ-
ent combinations of roll, pitch, and yaw pointing errors is presented in
54
1700
1600
LATITUDE, de9
010
1500
1400 30
1300
1200 0
1100
1000 505
_OO
COO
7OO
BOO _ 6050O
4OO
3OO
2OO
100
70
830 840 850 860
SLANT RANGE, km
870 880
Figure 23.
SEASAT SAR Doppler Center Frequency at Nominal Attitude versus
Target Slant Range at Four Sensor Positions (Descending Node)
55
-40 -30 -20 -10( I 1 I
m
ISODOPPLERCURVES
FD= -1500 (Hz)
0 10 20
- 240
I
I
-- 26O
I
l
--280
-- -- 300
--320
--340
- 360 ]500 "
(kin)
GROUNDRANGE
30 40 (kin)
I ALON OTRACK
CONSTANTSLANT RANGECONTOURS
-- -- 840 (kin)
--- 85O
B60
-- -- 870
-- -- 880
3000
Figure 24. SEASAT SAR Instantaneous Doppler
Over Target Surface for a Sensor
and Descending Node
Frequency ResponsePosition at 30" Orbit Anomaly
56
Figure 25. It is obvious from this figure that compensation must be provided
in a SEASATSARprocessor if the pitch and yaw pointing errors exceed 0. I to
0.2 degrees. The specifications on the SEASATattitude control system
permitted pointing errors of 0.5 degrees in each axis.f
For perfect antenna pointing, the SAR footprint aligns to the cross track
direction. The composite effects of the earth's rotation and the azimuth dis-
placement due to Doppler differential are now discussed for the case where
antenna pointing is perfect. Square ABCD on Figure 26 (a) is assumed as a
pattern in an image acquired by a high resolution real aperture radar
system. Because AB and CD are acquired at two different points in time, the
motion of the target field, with its direction marked by Ve, maps ABCD to an
area shaped by A'B'CD on the ground surface. This form of distortion is
inherent in a line-scan sensor, viewing a laterally moving subject. It is,
for example, observed in imagery obtained by the multi-spectral scanner
carried on the LANDSAT satellites. The angle 6 produced by this effect (see
Figure 26) can be approximated by the following expression:
V cosE sinse
Vs
with
cos_
(cos2_sin2_ + cos2_) I/2
_in-i sin_sin_
57
-40 -20,J I
LOOK ANGLE 17° - 230
(ROLL. PITCH, YAW): UNIT 1°
(-00
0
I, 220
20 40 (kin}I J
DISTANCE ALONG TRACK
(0+-)
k
(0-+)
(+00)360 (kin}
CROSS TRACK GROUNDDISTANCE FROM NADIR
Figure 25. Location of SEASAT SAR Footprint as a Functionof Attitude
58
(a)
A I _--
A l//t
I
c j
CROSS _TRACK
B"
B --/B'I
/ ALONGTRACK
//
....... IDO
D
B I
B //I
I
/
(c)
A' A B' B
ALONG\,, \
TRACK "/
CROSSTRAC K
(d)
A
\
\
¢,.\Ve',__
C'
\\\\
D'
Figure 26. Geometric Distortion of IDP Processed SEASAT SAR
Imaging and Distortion Due to Earth Rotation in a
Line Scan Imaging System
59
where
and
Ve
Vs
S
is the surface speed of earth rotation at equator
is the surface speed of the subsatellite point
is the target latitude
is the angle between the satellite velocity vectorand the earth's rotation velocity vector at the sub-satellite point
is the angle of orbit inclination (- I08°)
is the latitude of the subsatellite point
For a target at a latitude of 40° north, the angle _I is approximately
2.8 degrees.
In the SEASATSARsystem, the radar echo data must go through synthetic
aperture processing to form high resolution imagery. The range walk compen-
sation implemented in the IDP skews the echo data block to partially offset
range walk present in the data. The skewedecho data block is indicated by
A'B'CD on Figure 26 (b). The range walk compensation is implemented in dis-
crete angular steps. Typical values of this skew compensation angle _ are
approximately 0.9, 1.8, and 3.6 degrees. They correspond, respectively, to
adjustments of 16, 32, and 64 output plxels per every 1024 pixels in
azimuth. The adjustments compensatefor Doppler center frequency offsets of
360, 720, and 1440 Hz, respectively. This method of data block skewing to
compensate for range walk produces stepped edges in the processed imagery. An
example of such an edge is depicted in Figure 27. This figure illustrates the
boundary of the I00 km x 100 km SEASAT SAR image shown in Figure 11.
60
CROSSTRACK
ALONG DOWN SKEWING OF THE BLOCK
TRACK AS SHOWN HERE CORRESPONDS
TO THE RANGE WALK COMPENSA-
TION FOR POSITIVE DOPPLER
CENTER FREQUENCY (APPROACH-
ING TARGETS) THAT WASAPPLIED IN THE CORRELATION
PROCESS
I fL I
| , |
Figure 27. Image Data Block Skewing as a Result of
Range Walk Compensation
61
The IDP azimuth processing uses a single Doppler frequency value for the
entire 100 km swath width SEASATSARprocessing. The nature of the processing
is such that it enhances a target which has a response that accurately matches
the reference function• Whena single Doppler frequency value is used over
the entire swath width, the resultant image lines are aligned with
corresponding "isodoppler" curves of that frequency on the ground surface.
The motion of the earth together with the surface curvature causes the Doppler
frequency response to vary along the cross-track direction. The result is
that isodoppler curves are in general at a tilted angle relative to the cross-
track direction as shownin Figure 24. To avoid complexity in the image
recording process, the output image lines are given a cross-track orien-
tation. Consequently, a geometric skew distortion is introduced into the
images. This effect is described by angle 8 and the further skewed parallelo-
gramA'B'CD' in Figure 26 (b).
The angle B can be approximately evaluated in the following manner. The
difference between the assumedDoppler frequency and its true value is called
the Doppler mismatch and denoted by the symbol Af. The associated position
displacement in azimuth, Ax, can be approximated using the following formula:
AfAx = --V
• s
f
where f is the instantaneous rate of Doppler frequency change at the center of
beam, and Vs is the speed of the subsatellite point. This exPression is
derived from the correlation of twolinear FM chirp pulses of the same chirp
rate f, but differing in their center frequency by _f. Another simpler way to
analyze the problem is based on the familiar aircraft SAR equations [3], i.e.
62
phase delay
Doppler frequency f
2_x2 2nV2t 2
_R %R
2vx 2V2t
XR _R
• 2V 2 2VAx 2V2At
frequency rate gf = _R , from gf = _---R-_ _R
A simple manipulation of the last two expressions also leads to the formula
given previously.
At a latitude of 40 ° north, a typical value of Af which measures Doppler
differential between near and far swath positions is 200 Hz. (Compared to the
Doppler bandwidth of approximately 1300 Hz, a mismatch of + 100 Hz produces a
very slight degradation in the image SNR.) The values of f and V s are
approximately 520 Hz/sec and 6600 m/sec, respectively. The relative displace-
ment in azimuth, _X, is roughly 2.5 km from near to far ends over the swath
width.
The azimuth skewing angle _ is expressed as
AX
W
where W is the i00 km swath width.
approximately 1.5 degrees.
i
With AX being 2.5 km, the B angle is
The true composite skew angle is roughly the sum of the range walk compen-
sation angle Y and doppler mismatch skew $. The distortion of a SEASAT SAR
imagery acquired during an ascending node is opposite to that acquired in a
63
descending node as described in Figure 26 (a) and 26 (b). Figure 26 (c) and
26 (d) illustrate such effect corresponding to SARimagery obtained in an
ascending node. Note that the discussion given here does not include the
effects associated with extraneous yawand pitch deviations. It is also noted
that angle Y can be chosen such that the sumY and _ is about equal to the
inherent earth rotation effect 6. In this case the processed imagery is
relatively free from the distortion due to the natural earth rotation
effect. A subsequent slant range to ground range conversion can be performed
to provide a good geometric representation for SEASATSARimagery. Sucha
first Order geometric compensation schemeis intended to be investigated in
the near future.
!
5.1.6 Other SEASAT SAR Sensor Related Geometric Distortions
A change in the apparent geometric location of a target can occur if the SAR
sensor changes its parameters during its data acquisition period. Changes in
the radar PRF or the position of the echo sampling window affect the geometric
distortion most. A common result is that one part of the image is totally
displaced in range from another part. Such distortion was observed in a few
of the processed images. However, the transition reglon does not exhibit a
sharp discontinuity in the scene associated with the PRF change. Observation
of this distortion was made based on noticing dislocation of known target
patterns on the ground surface.
5.2 Radiometric Characteristics
The apparent brightness of a target in SAR imagery exhibits certain character-
istics which are very much different from that of an optically sensed
64 i
image. The first two characteristics treated in the following discussion,
namely the Rayleigh speckle and random noise, are inherent to SAR imaging and
appear in almost all the SAR fmage scenes. The later three characteristics,
namely the pulse compresslon sidelobes, the ambigufty responses, and the weak
signal suppression, are dependent upon the design of the radar system. They
are not apparent in most of the SEASAT SAR imagery. They are observed only in
images which contain extremely bright targets. The purpose of including such
unusual responses into this discussion is to provide the basic information and
typical examples to enable proper discrimination between them and other
natural causes in the SEASAT-A SAR imagery. It is also observed that target
response depends greatly on the SAR viewing geometry, especially for a struc-
tured target pattern such as that of city blocks. An example is shown in the
two images in Figure 33.
5.2.1 Rayleish Speckle Effect
The Rayleigh speckle effect [I0, 11] is particularly apparent in coherent
imaging of an objective having a surface roughness comparable to the
wavelength of the light source. The effect is caused by interference of light
scattering from those scatterers located more or less uniformly but randomly
over the target surface. For a large number of scatterers, the resultant
amplitude received by the sensor is random, and the resultant amplitude
response is best characterized by the Rayleigh distribution. The two
orthogonal - I,Q, - signal components prior to the amplitude detection exhibit
a normal distribution as a result of the central limit theorem which applies
to the sum of the large number of random inputs. The corresponding intensity
response, which is a square of the amplitude, follows the exponential
distribution. That is, for an expected intensity response I, the detected
65
strength I after synthetic aperture processing has the following probability
density function:
P(1) = exp(-I/l)
Note that the standard deviation of this distribution equals the mean value
I. Other characteristics of these distribution functions can be found in many
mathematical statistics references. The interference pattern may vary as a
function of the perspective viewing angle along-track and the radar carrier
frequency. Independent measurements can be made by changing such
parameters. The idea of multlple-look processing of SAR data is to obtain
such independent measurements of each pixel and to average them to reduce the
random variations. Another approach is to assume that the surface reflec-
tivity varies rather slowly relative to the dimension of a pixel. It is
therefore possible to reduce the speckle variation by averaging adjacent image
pixels.
The IDP produced SEASAT-A SAR imagery incorporates a 25m resolution, 4 looks
processing. Each look is produced from one quarter of the available azimuth
spectral band (coherent processing using full spectral band results in a
single-look image with an azimuth resolution of 6 meters.) In normal IDP
processing the azimuth resolution of each look is 25m meters. The multiple-
look averaging is done in the intensity domain. It was determined in a prev-
ious study that intensity averaging provides a more accurate representation of
the target reflectlvlty.
66
The four look averaging reduces the intensity variation by a factor of 2 (the
square root of the numberof looks). The remaining fluctuation is still
high. An example of such "speckley" variation is shown in Figure 28 (a) where
a section of the 25 m, 4 look SEASAT-ASARimagery over agricultural fields in
Imperial Valley, California, is illustrated. The speckle variation within
each field is clearly visible. Figure 28 (b) shows an image of the samearea
obtained by averaging the pixels shownin the previous picture. The speckle
variation is reduced at an expense of degrading the spatial resolution of the
image. Somediscussions on the effects of speckle on image spectra are given
in [12]. A copy of this reference is also given in Appendix B.
Radar backscattering from sea surface mayalso fit the Rayleigh statistics
model. This is because the short wavelength waves which typically ride on the
swells represent random scatters on the ocean surface.
Rayleigh speckle is observed over targets that have a surface roughness com-
parable or greater than the radar wavelength. For certain targets where their
surfaces are relatively smooth and are approximately perpendicular to the
direction of incident wave propagation, the intensity variations over the
looks are small. Such targets usually exhibit very strong reflectivity and
are generally referred to as "specular" targets.
5.2.2 Thermal Noise Effect
The detected signal power in a SAR imagery has two components. They are the
radar RF power reflected from the target area, and the noise power introduced
by the SAR sensor, the telemetry link, and the processing system. In an
intensity representation of the correlated imagery, these two components are
67
Figure 28.Speckle Reduction Through Image Smoothing
(a) A 25m Resolution Image Over Imperial Valley, California
(b) A Smoothed Lower Resolution (Approximately 60m)
Rendition of the Same Area as Shown on (a)
68
additive. The probability density function of the pixel intensity still
follows the exponential distribution as described in the previous section,
except that the mean intensity, I, is now expressed as the sum of radar RF
reflection and noise [13, 14].
I = S+N
The mean reflected signal power S depends greatly on the target reflectivity
over a resolution element, and the mean noise power N is modeled as a constant
for all the pixels.
As the number of looks increases, the noise component in a pixel approaches
the constant value N. By knowing the signal-to-noise ratio (SNR) of the SAR
system (SNR of SEASAT SAR is approximately i0 dB) and the system gain profile,
the value of N is readily estimated. The expression which relates N to the
SNR and the mean intensity P over the entire image is given below:
N = P/(SNR + I)
where
P " S+N
For large number of looks, this value of N can be subtracted from all the
pixel values to achieve a better representation of the target response. It is
also meaningful to include other interferences such as sidelobe and ambiguity
69
level (to be discussed in next subsections) into the noise calculation, and to
subtract these from the multiple look imagery.
5.2.3 Pulse Compression Sidelobes
The SEASAT-A SAR sensor adopted a linear FM (frequency modulation) chirp
waveform to encode its transmitted pulses° The chirp waveform has a time
duration of 33.8 microseconds and a bandwidth of 19 MHzo Compression of this
FM chirp waveform results in a sinc function - sin(x)/x - form of impulse
response [15] in the range dimension. The point target response in the
azimuth dimension can also be closely approximated by the linear FM chirp
waveform. Synthetic aperture correlation results in an impulse very similar
to that of the range pulse compression. The detected impulse waveform resem-
bles the square of sinc function. The response function outside the first
nulls of the mainlobe are referred to as sidelobes. These sidelobes interfere
with the target response at other pixel positions° The result is a loss of
detectability for weaker targets which are located very close to a strong
one. The sum of sidelobe interferences at any pixel position due to other
targets in the surrounding area can be modeled by calculating the mean energy
in all sidelobes. The ratio of the sidelobe response level (which may also
include the ambiguities - 5.2.4) to that in the main-lobe is called the
integrated-side-lobe ratio (ISLR). This ISLR is often used to quantify the
level of target induced interference in a radar image dominated by distributed
targets.
It is clear that a lower sidelobe response improves the image quality. For a
linear FM waveform_ it may be necessary to apply amplitude weighting to the
correlation reference function to reduce the sidelobe level. The weighting
70
applied in the IDP processing is very moderate (cosine square plus 0.5 pedes-
tal). The technique is to preserve the 25mspatial resolution of the corre-
lated images as muchas possible (amplitude weighting in the chirp waveform
broadens the malnlobe of the correlated impulse). The range peak sidelobe in
IDP produced 25mresolution, four-look imagery was measured to be approx-
imately -20 dB, and the azimuth peak sidelobe was measured to be approximately
-9 dB. Strong sidelobes are usually observed near very bright targets. An
example is shownin Figure 29. The extremely bright point to the left of the
image is the 26mantenna in the NASAGoldstone DSN(Deep Space Network) Track-
ing station. Rangesidelobes are clearly observed. For weaker targets, such
as the radar corner reflectors placed in the Goldstone dry lake, the sidelobes
are less visible (also see Fig. 12).
5.2.4 Ambiguous Target_ResPonse Due To Antenna Sidelobes
The ambiguous target responses in correlated SEASAT SAR imagery are due mainly
to sidelobes in the antenna radiation patttern. Such response can further be
classified as the range ambiguities and azimuth ambiguities. The range ambi-
guity is caused by antenna sidelobes in the range direction. That is, echo
responses due to sidelobes for targets at a much closer or further distance
than those targets illuminated by the mainlobe can add to the mainlobe echo
response (due to periodic radar pulse transmission and echo reception) and
produce spurious targets in the correlated imagery.
In the azimuth dimension, the target Doppler spectrum corresponds to the
antenna response in that direction. The finite radar PRF (pulse repetition
frequency) sampling in azimuth results in the fold-over of the Doppler
71
72
spectral energy from the sidelobes into the mainlobe. This aliasing effect
produces ambiguous target response in the azimuth dimension.
The composite ambiguity function of the SEASAT-A SAR radar system is approx-
imately plotted in Figure 30. The upper center spot is the desired main res-
ponse of a point target of the SAR system. Those points located at a further
distance from the center point are ambiguous responses resulting from finite
radar PRF and extended antenna radiation pattern. The spatial locations of
these ambiguities can be precisely calculated according to the radar PRF value
and the pulse compression parameters. The approximate locations of the range
and azimuth ambiguities measured in the slant range and the along-track dis-
tances are also indicated in the figure.
The PRF ambiguities are in general very much weaker than the main response so
that they do not normally produce visible effect in a processed imagery.
_i However, exceptions due to strong targets do exist. An example of such<'
_ azimuth ambiguity is shown in Figure 31. The figure contains a section of the
SAR image containing New Orleans. The brighter features in the lake are
caused by the ambiguous responses of the very bright area approximately six
kilometers south east of the lake shore. Range ambiguities are more difficult
to verify, because the targets which cause those ambiguities are located
outside the imaging swath and cannot be referenced without using another
image.
5.2.5 Weak Siznal Suppression Effect
The effect of weak signal suppression in the detected radar signal obtained
through correlation processing corresponds to that of the response of a weak
73
ISODOPPLER Fd - 1647
SWATH Rg _ 292 km_1[- 190 km/ GROUND
SURFACE
,,,|
Fd + 1647Fd
145 kmGROUNDSURFACE
RgGROUNDRAN GE
ALO NG-TRAC KDISTANCE
EQUAL SLANT RANGEIR =_ 860kin
S
R + 91 kmS
R + 182 kmS
Figure 30. SEASAT SAR Range and Azimuth Ambiguity Function
for the 1645 Hz Pulse Repetition Frequency
74
Figure 31. Azimuth Ambuiguity Induced False Features inMiddle of Lake Pontchartrain, New Orleans
75
L_°r°d°c°__r°__Ii best availabJe copy.
target which appears to be suppressed in the presence of a very bright target
in its close vicinity. Further, the suppression is stonger as the distance
between a weak target and the bright main target is reduced. This effect
causes a nonlinear loss (more for weaker targets) in detectability due to
saturation (due to limited dynamic range) in the uncorrelated radar signals.
In addition to a possible saturation in the radar data acquisition system,
signal saturation can also occur in IDP processing due to finite bit data
quantization in two intermediate steps. One step is after the range corre-
lation, where each of the complex data words is quantized into 16 bits with
8 bits for each real or imaginary component. The other step is after the
azimuth forward Fourier transform. The spectral data of the 8,8 bits range
correlated data are again quantized into the same number of integer bits. The
gains of the range correlation and the azimuth forward transform (the latter
is set to have a unit gain) are designed so that only a very small percentage
of the data will be saturated (assuming there is no substantial saturation in
the input data).
Revolution 1291 was found to have strong saturation in the digitally recorded
raw data. The weak signal suppression effect is observed in the imagery
processed from that revolution. An example is shown in Figure 32. This
picture is an enlarged section of the image shown in Figure 11. The horizontal
linear feature in the middle of the picture is the Santa Ana River. Freeways
in this area are shown as major dark lines. The dark bands next to several
bright features right above the Santa Ana RiSer are oriented in the azimuth
dimension, and are apparently due to the weak signal suppression effect. It is
possible to conclude that this suppression is due to saturation in the input
raw data, because the same area as viewed in another SEASAT revolution
I
76
Figure 32. Azimuth Near Signal Suppression Effect Shown
Near the Center of the Image
77
(Rev. 351) does not exhibit this weak signal suppression effect. It is also
possible that a change in the SARpass orientation (Revs. 351 and 1291 are of
ascending and descending nodes, respectively) maychange the apparent reflec-
tivity of those bright target features (5.2.6), and therefore reduce the
severity of the weak signal suppression effect. The IDP range and the azimuth
correlation gains are 27 dB and 25 dB respectively. For the example shown in
Figure 29, the suppression effect in azimuth indicates that the contrast in
target reflectivity wasmuchgreater than 25 dB but was less than 52 dB. If
the contrast becomesmuchgreater than 52 dB, the suppression effect may
becomenoticeable in both range and azimuth dimensions.
5.2.6 Target Reflectivity and Radar Aspect Angle
The apparent brightness of a target in SAR imaging depends not only on the
target electromagnetic characteristics but also on the geometry of the target
structure and the associated radar viewing angle. The latter factor may
become very significant for an area with man-made structures where the radar
targets or scatterers are geometrically oriented. An example is shown in
Figure 33. A section of the image taken near the upper left corner of Figure
29 of revolution No. 1291, a descending node, is displayed in Figure 33 (a).
The same area as seen in an imagery acquired during Rev. 351, an ascending
node, is shown in Figure 33 (b). The change in brightness in the central area
of the scene is primarily attributed to the different radar aspect angle. The
satellite path of Rev. 351 was approximately parallel to the street blocks.
The perpendicular faces of building structures to ground surface can induce a
strong radar reflection very much similar to that which occurs over radar
corner reflectors.
78
Figure 33. Effect of Radar Aspect Angle to the Apparent Target
Reflectlvity Over Cultural Targets
79
6. SUMMARY
This document provides a brief description of the Interim Digital SARPro-
cessor and the characteristics of SEASATSARimagery. It is hoped that these
engineering data are useful to the interpretation and analysis of the IDP
produced SEASATSARimagery. Performance evalution of the IDP processed
imagery is a continuing activity. It is also planned to modify the existing
software in the near future to improve the sidelobe performance and to provide
better radiometric calibration.
SEASATSARimagery in general demonstrates the concept and feasibility of
microwave remote sensing from a spaceborne platform. The extended surface
coverage, the near uniform radar angle of incidence, and the fact that the
sensor operation is relatively free from perturbation, are indeed advantageous
over conventional airborne SARsystems. Several phenomenawhich are often
obscure in airborne SARimagery are very clearly observed in SEASATSAR
images. With further efforts in the SEASATSARimage analysis, a muchbetter
definition than that which is currently available is anticipated.
80
REFERENCES
Io
.
.
.
1
o
R. Jordan, and B. Huneycutt, "SEASAT-A Synthetic Aperture Radar
Performance," IEEE 1979 International Conference on Communication Record,
Vol. 3, pp. 52.2.1-52.2.5, June 1979.
L. J. Cutrona, Synthetic Aperture Radar, Radar Handbook. M. I. Skolnik,
ed. McGraw-Hill, New York, 1970, Chapter 23.
E. N. Leith, "Quasi-Holographic Techniques in the Microwave Region."
IEEE Proc., Vol. 59, No. 9, pp. 1305-1318, Sept. 1971.
K. Tomiyasu, "Tutorial Review of Synthetic-Aperture Radar (SAR) with
Applications to Imaging of the Ocean Surface." Proceedings of the IEEE,
Vol. 66, No. 5, pp. 563-583, May 1978.
C. Wu, "A Digital System to Produce Imagery from SAR Data." Proceedings
of the AIAA Systems Design Driven by Sensors Conference, Paper No.
76-968. Pasadena, California, October 1976.
C. Wu, " A Digital Fast Correlation Approach to Produce SEASAT SAR
Imagery," Proceedings of the IEEE 1980 International Radar Conference,
5
.
Washington, D. C., April 1980.
W. E. Brown, Jr., B. Holt, and M. Strommen, "Catalog of SEASAT Experiment
Sites and Imaging Radar Requested (Preliminary Edition)", Jet Propulsion
Laboratory, February 1979. (JPL internal document.)
S. S. Rifman and D. M. McKinnon, Evaluation of Digital Correction
Technigues for ERTS Images, TRW Corporation Final Report, TRW
20634-6003-TU-00, NASA Goddard Space Flight Center, Greenbelt, Maryland,
March 1974.
81
9. W.E. Brown, Jr., C. Elachi, R. L. Jordan, A° Laderman, and T. W.
Thompson,Planetary Imaging Radar .Stud_, pp. 4.15-4.23, TR701-145, Jet
Propulsion Laboratory, Caltech, June, 1972 (JPL internal document).
10. R. L. Mitchell, "Models of Extended Targets and their Coherent Radar
Images," IEEE Proc. Vol. 62, No. 6, pp. 754-758, June 1974.
II. N. George, "Speckle," Optics News, pp. 14-22, January 1976.
12. C. Wu, "A Derivation of the Statistical Characteristics of SAR Imagery
Data," Proceedings of the Third European Space Agency SAR Image Quality
Workshop, December 1980, Frascati, Italy.
13. R. G. Lipes and S. A. Butman, "Bandwidth Compression of Synthetic
Aperture Radar Imagery By Quantization of Raw Radar Data," Proceedings
of the SPIE, U01. 119, pp. 107-114, August 1977.
14. C. Wu, O_timal Sampling and Quantization of Synthetic Aperture Radar
Signals, JPL Publication 78-41, Jet Propulsion Laboratory, California
Institute of Technology, Pasadena, California, 1978.
15. C. E. Cook and M. Bernfeld, Radar Signals, pp. 130-136, Academic Press,
New York, London, 1967.
82
APPENDIXA
A Digital SystemTo Produce Imagery From SARData
C, Wu
A-I
A DIGITAL APPROACH TO PRODUCE IMAGERY
FROM SAR DATA _;
C. Wu
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, California, 91103
Abstract
This paper describes a digital processing
algorithm and its associated system design for
producing image s from Synthetic Ape rture Rada r
(SAR) data. The proposed system uses the Fast
Fourier Transform (FFT) approach to perform the
two-dimensional correlation process. The range
migration problem, which is often a major obsta-
cle to efficient processing, can be alleviated by
approximating the locus of echoes from a point
target by several linear segments. SAR data cor-
responding to each segment is correlated sepa-
rately, and the results are coherently summed to
produce full-resolution images. This processing
approach exhibits greatly improved computation
efficiency relative to conventional digital pro-
ces sing methods. /
I. Introduction
It is known that for a given frequency the angu-
lar resolution capability of a radar system is di-
rectly determined by the aperture size of the radar
antenna. The idea of the synthetic aperture radar
(SAR) I,Z is to use coherent phase information in an
array of radar echoes to synthesize an effective
antenna aperture which is much larger than the
size of the physical antenna. This approach en-
ables high spatial resolution radar images to be
attained with practical size antenna.
In order to achieve the effective increase in
aperture, the raw radar returns gathered go
through processing to compress the dispersed re-
sponse from a point target. Typically, the pro-
cessing involves a correlation between the raw
echo data and the response function of point target.
Since the dimension of dispersion, i.e., the num-
ber of elements needed to be correlated to produce
an image point, is usually large,_ digital SAR-pro-
cessing is often characterized by a requirement
for a very large bulk memory to store the data to
be correlated. In addition to this requirement, in
most of the synthetic aperture radar applications,
the data rate from the radar receiver is very high.
To achieve real-time or near real-time process-
ing, the speed of data processing in an electronic
SAR processor must be very high to accommodate
the high input data rate. The requirement of
large bulk data storage and high data processing
speed in a digital SAR processor make it impor-
tant to properly design the memory management
and processing system. In this paper we shallfocus our attention to the methods for efficient
processing of SAR data into images.
Generally speaking, in a correlation process
the number of arithmetic operations is directly
proportional to the number of elements in the
reference function and the number of elements in
the final image. It is clear that the processing
efficiency (which is a factor in determining sys-
tem throughput) is directly related to the number
of arithmetic operations. If the computation
speed of an arithmetic unit is a limiting factor,
system throughput can be increased either by
implementing more arithmetic logic units to allow
parallel processing or by reducing the number of
arithmetic operations per image element by usingmore efficient processing algorithms. The cost
of processing associated with the former approach
increases linearly with the amount of parallelism
in the hardware. For the later approach, the
throughput can be increased at no increase in cost.
Special purpose hardware implementations may be
appropriate for operational applications, and the
optimum algorithm for such systems are influ-
enced by available hardware technology. To sup-
port SAR research activities, it is more practical
to accomplish processing in a general purpose
computer. Thus it is important to develop effi-
cient SAR processing algorithms for both special
purpose hardware processors and for general
purpose compute rs.
The two digital sAR processing approaches
used presently to reduce raw data into images can
be referred to as the time-domain correlation ap-
proach and the frequency domain FFT {fast fourier
transform) approach. The idea of time-domain
correlation is relatively straight forward. An im-
age element is processed by a direct reference
function multiplication and integration. The fre-
quency domain FFT approach uses the FFT to
transform echo signals into frequency spectra and
then filtering is applied to these spectra. The ba-
sic assumption of FFT approach is that the im-
pulse response function is linear and invariant
over the data block to be processed. Under this
assumption, the FFT approach does achieve the
optimality in the number of arithmetic operations
per each image element. However, there are sev-
eral difficulties in applying the FFT processing ap-
proach to SAR data which exhibit excessive range
curvature {wherein the locus of echoes from a
point target deviates from its linear approximationby more than one range resolution element in a
two-dimensional display of the SAR echo signals).
This problem is serious in some planned satellite
imaging radar applications, e.g., SEASAT-A.
This paper reports the results of an effort to for-
mulate a computationally efficient FFT approach
which performs full synthetic aperture processing
of SAR data containing significant range curvature.
In the following discussion, a brief description
of the analytical background of FFT filtering
This pa_er presents the results of one phase of research carried out at the Jet Propulsion Laboratory,
California Institute of Technology, under Contract No. iNAS7-100, sponsored by the National Aero-
nautics and Space Administration.
A-2
approach for SAR data processing will be given,
and then the difficulty of efficiently appIykng the
FFT approach for processing SAR data with signif-
icant range curvature will be discussed. Major
emphasis is placed on rigorous description of the
method for efficient SAR processing in conjuction
with the correction for range curvature.
If. Digital Processing of SAR Data
Using the FFT Approach
Assume a wave form cos (_t - _(t)) is radiated
from a synthetic aperture radar, where w/2_ is
the frequency of the coherent carrier and-oh(t) is
the pulse coding form. A point target a distance
r 1 from the aircraft returns the signal:
Zrl/ Zrl nT)I (1)s(t,- coswhe r e
is the radar cross section (reflectivity) of the
target
c is the speed of light
and
T is the pulse repetition interval
Since the pulse repetition rate of an SAR is such
that a point target on the surface is "interrogated"
many times as it passes through the beamwidth of
the radar antenna, the ensemble of returned echo
signals can be assembled into a rectangular dis-
play _-ith elapsed time (range, r) along one axis
and along flight position (azimuth, x) along the
other axis. The assembled returns from a point
target is also two-dimensional. For a point target
of unit reflectivity located at (Xo, ro), the two-
dimensional impulse response at baseband output
can be written as
h (x - x 0, r - ro) =
exp j {-_--_-- _c r0) " c (rl "
for (x, r) eR
0
otherwise(z)
whe re
Ir z x0 )zrl = 0 + (x - (3)
R, the region of responsiveness, is determined by
the pulse duration and the pattern of the antenna
beam. Eq. Z can also be expressed as
h (x, r)= h I (x, r) ® h 2 (x, r) (4)
where "® " stands for convolution, and
hi(X' r)=Ifor (X, rl eR. 1
{o_.othe rwtse (5)
A-3
hz ix, r) :
Zr
5(x) exp j ¢(c} for (x, r) cR Z
0 otherwise (6)
R.,, R.6 are.the region of responsiveness, for h l.and
hz, respectively. The 5-function in h i and h z ts
D]rac's delta function. In terms of physical
meanings, h I is the point target return corres-'
ponding to an "infinitesimally narrow" transmitted
pulse, and h 2 has the waveform of the actual trans-
mitted pulse.
Eqs. Z to 6 summarized the response from an
ideal point target. For an extended target,
• (x, r), the response f(x, r} can be written as:
f(x, r): _(x, r) @ h(x, r) (7)
Substituting Eq. 4 into above, we have
f(x,r)= _(x, r) ® [hi<x, r) ® ha(x, r)_ (S)
It can be shown that Eq. 8 has the following
alternative form
f(x, r)=[_(x, r)® hl(X, r)?, h2(x, r)(9)
The above equation shows that the echo rett/rns,
fix, r), can be expressed as the original target
field or(x, r) conyolved sequentially with two
impulse responses h I and h 2. It is therefore
clear that the targets _(x, r) can be reconstructed
by sequentially correlating the return signal
f(x, r)bythe responses hz(x, r)and hl(x, r).
Eq, 6 shows that hz(x, r) is one-dimensional (in
the range direction) wave form identical to the
transmitted pulses. ,Correlation with respect to
h z thus is a conventional one-dimensional corre-
lation operation with a linear fixed reference func-
tion. One can therefore use a FFT approach to
perform the range correlation. The correlation
with respect to hl(X, r) is generally referred to
as the azimuth correlation. According to EQ. 5,
the locus of hl(X, r) is a two-dimensional curve.
The dispersion in the x-direction is determined bythe radar antenna beam width or the size of the
synthetic aperture. In the r-directlon, the change
is due to the varying of distance between target and
sensor as defined by Eq. 5. Indeed, the locus of a
point target (measured by the delta function and the
range of x) represents a section of a hyperbolic
curve as defined by Eq. 3.
In many imaging radar applications, the geo-
metric factors are such that the locus can be
satisfactorily approximated by a straight line.
That is, the maximum deviation of the true locus
from the straight line approximation does not
exceed half of a range resolution element. In such
cases, the correlation Of hl(x, r) can be viewed as
a one-dimensional correlation problem. The FFT
approach can again be used if the size of the data
block is selected such that the azimuth reference
function h l(x, r) is approximately unchanged over
the data block to be processed.
In the above discussion, we have shown that if
the range curvature .is insignificant, the SAR
correlation process can be performed by two-
sequentiaIFFT correlation operations. A block
diagram of such an approach is shown in Figure I.
The raw data are first range correlated to
compress the phase coded pulse into a muchnarrower pulse. The range processed data blocksare stored, and then retrieved along the direction
of azimuth correlation. The transfer function ofthe azimuth filtering which is derived fromhl(x, r) needs to be constantly updated for targets
at different range r 0. This is because the waveform of the reference function hl(x0 r) changes
with respect to r 0. Because the azimuth referencefunction hl(x, r) is target range dependent, optc:i-
real processing requires range correlation be per-formed prior to the azimuth correlation. Other-wise the azimuth reference function is not unique
for a given range r 0.
RAW DATA
FFI
I •
J
RANGETRANSFER
FUNCTION
I AZIMUTHTRANSFER
FUNCTION
r
r
[..°11CORNER
TURNING
J FFT
!!
IMAGES
Figure 1. _ System Block Diagram of the
FFT SAR Data Processing Approach
Numerous SAK images have been producedfrom simulated and real SAR data using the FFTapproach discussed in this section. Results areall very satisfactory.
III. FFT SAR Processing with RangeCurvature Correction
In some SAR applications, there is a significantrange curvature associated with the locus of theechoes from a point target. In such cases it is nolonger adequate to use an approximate azimuth
reference function which lies on a straight line.Consequently, one cannot use the one-dimensional
FFT fast correlation approach for azimuth pro-cessing as discussed in the previous section. Weshall now describe a more general approach to theSAR azimuth processing to accommodate the cor-rection for range curvature. (The method pre-viously described is a special case of this moregeneral method. )
One approach to processing SAR data containingrange curvature would be to use a two-dimensionalFFT filtering, but this approach results in a highdegree of computational complexity. An alterna-
tive approach proposed here involves using an en-semble of functions lying along parallel straightline segments to approximate the curved azimuth
reference functions. The computational efficiencyof this approach is nearly the same as that of_the
processing method discussed in the previous sec-tion for SAR data having an azimuth referencefunction which lies along a straight line. With the
line segments approximation {or assuming that theazimuth response which has width in real cases issampled along a set of parallel lines), the azimuth
reference function, hl{x, r), can be representedas follows:
n
h I (x, r) = Z gi (x, r)
i=l
(10)
where
gi (x, r) _-
(r-di)_xPJ,'_ V'0 ' "':
_i -< x < ai+ 1¢
[ otherwise (1 1)
The distance of the ith line segment from the line
corresponding to r 0 is denoted by d i. The azi-muth coordinate corresponding to the start of the
ith line segment is denoted by a i. Fig. 2 graphi-cally illustrates this method of obtaining an ap-proximate azimuth reference function. Note thedirection of the parallel lines can be chosen alongthe azimuth (perpendicular to range delay) direc-tion or along a tilted direction to reduce the num-
ber of segments.
ia I a2 a3 a4 a5 °6 °7
Figure 2. A Curved Range Delay History(Solid Curve) and its Piecewise
Approximation (Dashed Line s)
Using Eq, 10, range correlated SAR data in sam-pled data domain can be represented as follows:
n
s(x, r)=or(x, r) @ --_ gi(x, r) (lZ)i:l
A s suming the approximated a zimuth reference func -
tion in Eq. 11, the required azimuth processing canbe accomplished with one-dimensional FFT pro-cessing. The theoretical background of this ap-proachwillbe discussed Lnthe following paragraphs.
For a constant r0, the fourier transform ofEq. IZ gives
A-4
s(u, v)= r_Iu, v).i._ oiIu, v) (13>L,--i A
The symbols S,_, and G i represent the transfor-
mations of functions s, _ , and gi, respectively,
and u, v represent the variables in the trans-
formed domain.
Eq. 13 implies that
n
Gi::: (u,v)
_(u, v) = S(u, v) i=l (14)
o,u.where ')".,:' denotes complex conjugate. The Fourier
transform of gi (x,r} according to Eq. 11 is
G i(u, v) = h i(u) exp (Jdiv) (15)
where hi(u) is the Fourier transform of Lhe expo-nential factor of gi, which depends on variable x
only. A close approximation of hi(u ) can bederived by using the fact that thedistance from
target to sensor is normally much greater than
the size of the synthetic aperture. This implies
that distance r 1 of Eq. 3 can be approximated by:
(x - x0 )z
r1-r o + ar° if%>> [x - Xoi (16)
An approximate expression for hi(u ) can there-fore be written as follows"
x,}i f ai+1' exp r0 expO )L C
- _b exp j(_'_ rO u ?') [i (u)
whe re
_(= expj (- Z_-_-r 0))is a constant phase
factor, and
(17)
_.y )dyIi (u) = /yi yi÷l exp (-j _ 2
whe re
r 2_ uYi = wcr 0
(18)
The integral Ii(u ) shown above is a complex
Fresnel integral. The magnitude of I i can beapproximated by the method of "Cornuls Spiral ''3.
Accordzng to Eq. 17, the magnitude of hi(u) is pro-
portional to that of If(u). Fig. 3 contains an
approximate characterization of the magnitude of
hi(u ) with respect to u. Each hi(u ) is shown to
occupy a distinct frequency band in the transformed
domain. This can also be explained by the fact
that the phase of hl(x, r) with r 1 approximated by
Eq. 16 resembles the phase of a linear FM chirp
wave form; and there is a unique correspondence
between each time interval in a linear frequencysweep "and a frequency band in the swecp bandwidth.
lh,l lh21151 lh,l lh l hl.... .... -q
, I , I Y I I
_l lit 3 i%'- .ll'l _ /i_, / Ill _%-
u I u2 u3 u4 u5 u6 u7
Figure 3. Amplitude Spectra of Segmented
Azimuth Reference Function
When the above results are applied to Eq. 15,
it is clear that Gi(u, v) also exhibits bandpass
characteristics. The amplitudes of Gi(u , v) are
approximately equal in the corresponding bands,
thus the denominator of Eq. 14 is nearly constant
throughout the total bandwidth of the azimuth
response. We will use K to denote the value of
the denominator in Eq. 14. Then, substituting
Eq. 15) into Eq. 14, we have:
fI(u, v)=I_ S(u, v)h;:: (u)exp (-j div )i=l
(19)
Inverse Fourier transform of the above equation
with respect to v yields:
I n
fI'(u, r} =--_-_-_ S* (u) r) hi:' (u) 6 (r d i)i=l
(Z0)
The quantity _'(u, r) is the filtered azimuth
spectrum. An SAR image will be formed by an
inverse transform of P.'(u, r) with respect to u.
For a fixed r in Eq. 20, the filtered line
spectrum is a superposition of a number of fil-
tered segments of line spectra. The range loca-
tion of each segment relative to r 0 is specified
by the value of d i. In processing a line of image,the superposition of spectral bands can be effi-
ciently implemented by a delay and registrationoperation. The method is 'described as follows:
After range processing (if the radiated pulse is
coded), each azimuth line is transformed into
frequency domaLn. The line spectrum is sub-
divided into a number of bands each corresponds
to the spectrum band of gi(x, r). The first sev-
eral lines are transformed to enable a complete
composite spectral line to be formed by super-
positioning spectral bands from different lines
according to the delay specified by the value d i.
The composite spectrum is then multiplied by the
azimuth transfer function h["(u), inversely trans-
formed, and filtered to form a line of the SAR
image. Each additional azimuth line transforma-
tion enables formation and processing of the com-
posite spectrum corresponding to the next line of
the image. This process will continue until the
whole block of data is processed into image. A
diagram of this processing approach is shown in
Fig. 4. The azimuth locus in Fig. 2 is used in
this example. It is shown that the composite
spectrum of an image line is obtained by assem-
bling the appropriate segments from the spectraof several azimuth lines. To accommodate the
A-5
range dependent nature of the azimuth response
function, the values of ai and the azimuth refer-ence function must be periodically updated when
there is a significant change in range r0.
If the input radar returns are contaminated b'y
noise, various filtering functions, such as the
least mean square error filter, minimum SNRfilter, etc., can be used to optimize the quality ofthe processed image. It is also clear that imagedistortion associated with this particular approachcan be measured by the amount of overlap in the
spectra segments shown in Fig. 3. Increase thenumber of segments in azimuth reference func-tion results in the increase of spectrum overlap.Therefore, it is very much desirable to choose adirection for azimuth processing {direction of the
line segments) which minimizes the number ofline segments in approximating azimuth refer-ence functions.
_=_ LINE 1LINE 2LINE 3
_,__I LINE 4LINE 5
r, [ E, I--q
I- .... I i+l I
r-rBT_-T --'_I I i l i+ll 1
c_-rB.---r-"I i I i+h
SPECTRUM
ALONG LINE I
SPECTRUM
ALONG LINE 2
SPECTRUMALONG LINE 3
SPECTRUMALONG LINE 4
SPECTRUMI ALONG LINE 5
I SUPERPOSEDI SPECTRAL BANDSIAi 'Bi I Cil Di ] Ei ] Fi
u 1 u2 u3 u4 u 5 u6 . u 7
Figure 4. Delay and Coherent Superposition
of SAR Azimuth Spectral Bands
Some experimental results are described inthis paragraph, The inphase components of two
simulated azimuth responses of a point target areshown in Fig. 5(a) and 5(b). The response in
Fig. 5(_) lies along a straight horizontal line,whereas the curvature of the azimuth response inFig. 5(b) spans four range resolution elements.Fig. 5(c) and 5(d) show the images of a point tar-get which were reconstructed from the signalsassociated with Fig. 5{a) and 5(b), respectively,using the filtering method discussed in Section IT;that is, using an azimuth reference function whichlies along a single straight line. Because correc-
tion for range curvature is not applied, the degra-dation of resolution is clearly observed in theimage of Fig. 5(d). Fig. 5(e) shows the recon-
structed image produced from the signal in Fig.5(b) using the method described in this section.By comparing the reconstructed point images,-it
is clear that the pulse width in Fig. 5(e) which isassociated with the process with range curvature
correction is much narrower than that of the
image in Fig. 5(d). It is also noted that the image
in Fig. 5(e) is nearly identical to the image inFig. 5(c). This demonstrates the validity of theproposed SAR processing method using the FFT
filtering approach for applications involving rangecurvature.
(a) Straight
(b) Containing Range Curvature
(c) Straight, No Correction
(d) Curvature, No Correct;on
(e) Curvature, With Correction"
Figure 5. Simulated SAR Azimuth Responses
and Reconstructed Images
IV. Conclusion
The method for digital processing of SAR sig-nals into images discussed in this paper is shownto be a valid processing method both analyticallyand experimentally. Because the FFT processingapproach is computationally efficient, tliismethod
can be very useful in applications where arithmeticcomputation speed is a significant factor in deter-
mining the cost of accomplishing timely process-ing of SAR data. Future electronic processingtechniques are likely to be substantially influencedby advances in electronic device technology. Inthe near term, the proposed method provides an
at-tractive approach for digital SAR processingusing available digital electronic technology.
Acknowledgement
The author wishes to thank R.G. Piereson forhis many thoughtful suggestions made in the course
of this work. He also wishes to thank A.J. Spearfor his encouragement.
Reference
1) E.N. Leith, "Quasi-Holographic Techniquesin the Microwave Region, " IEEE Proc.,
Vol. 59, No. 9, pp. 1305-1318, Sept. 1971.
2) R.O. Harger, Synthetic Aperture RadarSystems, Theory and Design. New York: rAcademic Press, 1970,
3) F.A. Jenkins and H.E. White, Fundamentals
of Optics, Chapter 18, McGraw Hill,New York, 1957.
A-6
APPENDIX B
A Derivation of the Statistical Characteristics
of BAR Imagery Data
C. Wu
B-I
A DERIVATION OF THE STATISTICAL CHARACTERISTICS OF SAR IMAGERY DATA*
C. Wu
Jet Propulsion Laboratory
4800 Oak Grove Drive
Pasadena, California 91109, U.S.A.
ABSTRACT
SAR imagery contains speckle noise as a result of
coherent interference of radar echo and surface
scatterers. This speckle effect must be properly
treated in analyzing SAR image quality and image
processing techniques. This paper, based on the
Rayleigh scattering model, derives statistics of
SAR sensed measurement and their relationships to
the surface mean power reflectivity. These
derived quantities are useful in areas such as SAR
calibration and image processing.
I. INTRODUCTION
Synthetic Aperture Radar (SAR) is a coherent sen-
sing device. Images gathered by a SAR exhibit
extraneous granularity as a result of interference
between the coherent electromagnetic wave trans-
mitted by the radar and the scatterers on the
target surface. Such granularity also exists in a
laser illuminated scene and is commonly referred
to as speckle [I].
The speckle variation effect presents several
difficulties in utilizing SAR image data. The
prominent difficulties are the large variance in
the radiometric response of a SAR picture element
(pixei), and the altered spatial correlation pro-
perty relative to that of a scene obtained by an
incoherent sensor. This paper addresses several
basic statistical properties of the speckle effect
and the associated spatial correlation of SAR
image data. In the following sections, a review
of the Rayleigh speckle model will be given. The
Raylelgh statistics will then be used to derive
the correlation and spectral measures for SAR
image data. Application of those derived statis-
tics to SAR radiometric measures and image proces-
sing will also be discussed.
2. RAYLEIGH SPECKLE STATISTICS
SAR processed echo measurements contain both the
target amplitude and phase responses before the
intensity detection procedure. To analyze the
statistical property of the SAR measurement for a
pixel, we begin with a derivation of the relation-
ship between the SAR measurements and surface
reflectivity.
Assuming unity radar illumination at target sur-
face, the echo return from a point target with
amplitude reflectance c o can be expresse_ as:
j%(x,r)ec(X,r) = %(x,r)e (t)
where (x,r) denote spatial coordinates in azimuth
(along-track) and range (cross-track), respec-
tively, for the location of the point. ¢0 is thephase delay which is a function of radar wave-
length and path distance only and is independent
of _0" The target surface under SAR interrogation
can be considered as a continum of infinitesimally
small point scatterers. The radar system response
to a point at (x,r) extends a finite distance over
(x,r) as a consequence of the limited transmitter
and Doppler bandwidths. Let us denote the system
response function as U(x,r). In many published
articles [2], U(x,r) takes the sinx/x waveform in
both range and azimuth direction as a result of
compressing a signal with quadratic phase
history. The SAR processed echo measurement is a
linear superposition of responses from all the
infinitesimally small scatterers. An expression
for a SAR measurement can be written as:
f/ JCo(Y'Z)A(x,r) = o0(y,z)e u(y-x,z-r)
6F
dydz + n t = JJ u(y,z)c0(Y+X,z+r)
J¢0(Y+X,Z+r)
e dydz + n (2)t
where n_ denotes the random noise component intro-
duced by the radar system.
The e_c= SAR response function is another
topic. Indeed, one can combine U and eJ¢O as the
real SAR response, and leave o 0 as the only vari-
able which relates to target surface. The I/2
power value-of U(x;r) which is commonly referred
to as SAR r_solution/is generally much bigger than
radar wavelength. For this reason, the phase
factor $0 changes over many cycles in U(x,r), and
hence the sign of the real and imaginary parts of
this complex phase factor also alter many times in
U(x,r). The expected value of the real and imagi-
nary components of A(x,r) are, therefore, zero.
*This paper presents the results of one phase of research carried out at the Jet Propulsion Laboratory,
California Institute of Technology, under Contract No. NAS7-100, sponsored by the National Aeronautics
and Space Administration.
B-2
The centra[ limit theorem which applies to the
summation of large number of independent variables
of tile same distribution also suggests that the
values' of the real and imaginary components of
A(x,r) are of Gaussian distribution. These two
components share equal variance, Oa2 , which is
one-half of the mean power of A(x,r), i.e.
These two density functions are of Rayleigh and
exponential distributions, respectively. Suos_i-
luting Eq. 5b into above, the amplitude and inten-
sity discribucions of a SAR measurement in terms
of the target power reflectivity PO of Eq. 4 are:
2a
2a P0+Nt
2(x,r ) = i fA(a) = -- e (Sa)a 2 eIIA(x'r)12] Po+Nt
P
= 21 [f]U2(y,z)O02(y+x, z_r) fp(p) = P0+N---_Ie Po+Nt (8b)
dydz + Nt] (3)
where N t denotes the mean noise power. Reduction
of the quadriple inCegratlon to double integratlon
is possible because of the independent phase
factors which result in a Dirac _-function during
integration. The integral above represents the
sum of power reflectivity 002 weighted by the
square of SAR system response U(x,r). We denote
P0 to be the integral of power reflectivity, i.e.
P0(x,r) = ffU2(y,z)o02(y+x, z+r)dydz (4)
.The mean and variance of the real and imaginary
components of A(x,r), which are represented by a I
and a 2 respectively, are explicity expressed as:
" m = e{a ] = = 0 (ba)a 1 E[a2]
2- V[a ] = V[a2]°a I = _(P0(x,r) + Nt)(5b)
Note that a I and a 2 are statistically uncorrelated
quantities because cost0 and sln_ 0 are uncorre-
fated for a random $0 uniformly distributed over
(0,2_). The joint Gaussian density function
for a I and a 2 is:
1 -(a12_'a22)/2°a 2
f(al'a2) 2 e (6)2TO
a
The probability densities of two derived quanties:
a =_al 2 + a22
2 2
p = a I + a 2
which are the amplitude and intensity of A(x,r),
respectively, are [3]:
2a
220
fA(a )_ _ - a a2 e for a > 0 (7a)Oa
P
22c
fp(p)__ _ I a2 e for p _ 0 (7b)2_
a
The measurement A(x,r) or its power P(x,r) can
also be written as:
i/2 J¢o
A(x,r) = a P0 (x,r)e + n t (9)
P(x,r) Y(P0(x,r) + N t) (I0)
where a, 7 are two random variables of distribu-
tions similar to Eq. 7 with Oa 2 = I/2, i.e.
2
f(_) = 2ae -_ (lla)
f(y) = 'e -_ (llb)
and the moments of _, v are [3]:
and
E[an ]
• l.n/2
i . 3 .... n tb) for n odd
for n = 2K
(t2)
z[Yn] - n: (13)
3. THE STATISTICS AND CORRELATION
PROPERTIES OF $AR IMAGE DATA
Eq. I0 clearly indicates that =he target power
reflectivity can be estimated linearly from the
SAR measurement P(x,r). The exponentially distri-
buted random variable 7 causes uncertainty in _he
estimate (where we assume the mean system noise
power Nt is known). We provide a generalized
expression for the expected value of the n-th
power of P(x,r) using Eqs. IO and 13:
E[pn(x,r)] = n!(Po(x,r) * N t) (14)
Based on this equation the mean and variance as
well as higher order quantities of measurement
P(x,r) are readily obtained. These quantities can
be used to derive the estimation accuracy as a
function of sample size, number of looks, and
expected pixel power P0(x,r).
The auto-correlation of P(x,r) is expressed:
R n(Ax,gr) = E[pn(x,r) . pn(x+bx,r+5r)] (15)P
B-3
The statistic at Ex=0 and gr=O reduces to the
value given in Eq. 14. For &x and gr greater than
the spatial resolution defined bY u(x,r), the
factors are approximately independent, i.e.
R n(gx,gr) (n!)2E[(P (x,r) + N )np q t
x,Lr+Ar) + Nt)n ] (16)(Po(X+
The expected value for the quantity within the
square bracket can be evaluated based on the cor-
relation property of Po(x,r) which resembles the
reflectivity scene obtained by an incoherent sen-
sor. For n=l, the correlation reduces to:
po(&X,_r ) 2 (17)Rp(gX,&r) = R + 2PoN t + Nt
where PO is the mean scene reflectivity. Therelationship between auto-correlation of the
coherently sensed scene and the corresponding
value of an incoherently sensed scene is given
above. Estimation of Rp_ from Rp is possiblesince _ and N are quantities that can be
• tobtalne_ by knowing the general signal to noise
ratio of the SAR system.
The cross-correlation of two SAR image sets is
also examined here. Let P1(x,r) and P2(x,r)
denote two data sets. Since two independently
measured data sets have independent speckle vari-
ation, the cross-correlation is simply:
R_n_n(&x,Ar)_l_2 = (n')2RPoiPo2(&x,_r) (18)
4. SPECTRAL CHARACTERISTICS OF SAR IMAGE
4.1 Spectral components of SAR imagery
The y factor in Eq, lib varies from pixel to
pixel, hence the power measure can be written as;
P(x,r) = Y(x,r)(Po(x,r) + Nt) (19)
Fourier transform of (19) results in the following
expression:
!
S(u)v) = 7 Sy(u'v)_IS0(u'v) + NtB2_(u'v)]
S_(u,v)_)So(u,v) + NtSx(u,v)_ B2 (20)
where_stands for convolution and S) Sv, SO
represent th_ spectra of P, y, and PO, respec-tively, and _ is the size of the spectral domain.Define:
Y= i + t
Eq. 12 implies
Elt] = l
v[t] = I
(21)
(22a)
(22b)
The spectrum of T(x,r) therefore is:
Sy(u,v) = B2d(u,v) + St(u,v) (23)
where S t is the white spectrum of t(x,r) of
Eq. 21. Replacing Eq. 23 into Eq. 20, we obtain
1
S(u,v) = So(_,_ ) +_-_ so(_,v)_st(u,v)
_2Nt_(u,v) + NtSt(J,v) (24)
Due to the fact that spectrum St(u,v ) is white
Gaussian, the convolution in Eq. 24 also results
in a white spectrum NT(u,v). The four componentsof S(u,v) can De expressed as follows:
S(u,v) = So(u,v) • Ny(u,v) + B2Nt_(u,v )
• NtSt(u,v) (25)
S O is the spectrum of the scene reflectivity.
Ny is the white noise spectrum due to speckle
variation. B2Nt6(u,v) corresponds to the spectrum
of the unity biased mean floor of the thermal
noise power N t. NtSt(u,v) is the spectrum due to
the variation of thermal noise. Baaed on the
variance of t as given by Eq. 22, the following
relationships are valid:
ffis0(uv>i2=udv= ffis (u, )Imdudv (26)
and
2 = f/ Nt2 iSt(u,v)[2 du dv (27)B2Nt
The above equations indicate that spectral energy
is divided equally over the original spectra,
S O and B2Nt6(u,v ) , and rue parts due to speckle
variation NT and NtS t. A graphic representation
of the expected energy of those four components is
given in Fig. I. The level Ny and the delta-func-
tion B2N t components are introduced due to the
speckle effect to the scene P0(x,r) and the posi-
tive noise floor Nt in a detected image. Given
the signal to thermal noise of the SAR system, the
partitioning of these levels are readily
obtained. The sum of N y and NtS t amounts to half
of the total power in the SAR image data.
SPECTRAlINTENSITY
SSS SS (u,v)
B2N) 8(0,0)
=0
,//; / .27/ / / ////////////-,_
Figure i. Four .Major Components in a SAR
Image Spectrum
B-4
_.2 6pectral filterin< of SAR i_iagery
The p_rtition_g of spectral components of a SAR
i=age de,or!bed above provides a quantatttive
basis to fo_nulate effective spectral filtering
techniques for SAK image enhancement. Applying
the well known !east-mean-square filter:
go(u,v) So(u,v)H(u,v) = (28)
_(u,v3 _o(u,v) + _(u,_)
where a bar indlcates the mean of the quantities
it designates. The filtered spectrum is:
S(u,v) = E(u,v) • S(u,v) (29)
For SAR image, the composit noise spectrum con-
sists of the last three terms of the right hand
side of equation (25). These levels can be
estimated by knowing the system signal-to-noise
level as mentioned in the previous subsection.
4.3 Simulation results
Simulated SAR image_' of ocean wave scenes were
used to study the scene spectra and image resto-
ration. They are shown in Fig. 2. The top three
images are simulated wave patterns of three diff-
erent wave spectral characteristics. The
coherency of the periodic wave structure decreases
from left _o right. Speckle effect was applied to
these images with a multiplicative factor which
corresponds to _he mean of four independent expo-
nentially distributed data samples. They are
sho_. in the second row of the figure. This four
sample averaging of the speckle factor described
in Section II simulates four-look S_R image
data. Pictures in Fig. 3 are the power spectra of
the images shown in Fig. 2. Note that the speckle
effect produces a'uniform and noisy level over the
entire two-dimenslonal wave spectra. The top
three pictures of Fig. _ are the filtered wave
scene obtained by applying the least-mean-square
filtering to the speckly images in Fig. 2. The
intensity level is lower than that of the original
images aince the filters as specified by Eq. 2
will reduce the filtered image power by a factor
of at least 2. Another filtering approach was
experimented with a threshold to the scene
spectrum. All spectral data with magnitude
greater than a defined threshold will be retained
and the rest will be set to zero. The purpose of
this threshold is to keep only those dominant data
which correspond to the scene spectral response as
shown on Fig. I. The threshold filtered wave data
are shown in the lower part of Fig. 4. Results of
threshold filtering appear inferior to that
obtained by the least-square-filtering approach.
5. CONCLUSION
The statistics of SAR image data derived in this
paper provide the needed mathematical basis to
estimate SAR target reflectivity and scene spec-
tral responses. A number of SAR multlple-looks is
necessary to reduce the variation due to speckle
effect. The formulas described herein enable the
estimation of size of sample or number of looks
required to estimate those quantities of interest
within certain specified accuracy.
6. REFERENCES
If] J.C. Dainty, "An Introduction to Speckle,"
Proceedings of the Society of Photo-Optical
Instrumentation Engineers, Vol. 243,
Paper 243-01, July 1980.
[2] E. N. Leith, "Quasi-Holographic Techniques in
the Microwave Region," IEEE Proceedings,
_oi. 59, No. 9, pp 1305-1318, Sept. 1971.
[3] A. Papoulis, Probability, Random Variables_
and S_ochastlc Process, pp 194-195.
McGraw-H_il!_l, 1965.
(a)
(b)
Figure 2. (a) Simulated Oceanwave Reflectivity Patterns
(b) Simulated Four-Look SAR Imagery
B-5
(a)
(b)
Figure 3. (a) Spectra of the Ocean Reflectivity Patterns
(b) SFectra of the Simulated Four-Look SAR Imagery
(a)
Figure 4. (a) Least-Mean-Square Filtered Four-Look SAR Imagery
(b) Threshold Filtered Four-Look SAR Imagery
---B-6
APPENDIXC
SEASATSARRadiometric Calibration Considerations
B. Huneycutt
C-I
C.I INTRODUCTION
There are manysystem gain variations and non-linearities which introduce
significant uncertainties into the amplitude calibration of the SEASATSAR
system. Those calibration error sources which are not controllable by the SAR
system, such as atmospheric and ionospheric effects and Faraday rotation _ .
effects, will not be addressed herein. Those calibration error sources which
are controllable by the SARsystem are best described by considering the
various elements in the SARsystem as shownin Figure CI. Only those elements
in the SARsystem up to the recording of the SARdata upon the hiBh density
data record (HDDR)are considered in this report. These calibration
considerations are commonto both digital and optical processing of the SAR
data, both of which access the HDDRdata.
i
C-2
_-- c_
_ A
z
I--
J.J
c_
C_,-4
_J
U_
0*,-4
e_
,,.-4
_J-,-i
0
P_
C.J,
OJ
C-3
C.2 CALIBRATION CONSIDERATIONS OF TBE $ARTRANSMITTER AND ANTENNA POINTING
To obtain amplitude calibration of the SEASAT SAR system, the system gain
variations caused by the SAR transmitter and the SAR antenna pointing must be
considered. For instance, the peak output power of the SAR transmitter FM
pulse decreases typically by 0.6 dB with increasing transmitter temperature
over the nominal I0 minute SEASAT SAR pass°
The L-band signal is radiated by the SAR planar array antenna; for a uniform,
extended target, the return echo is amplitude modulated in the range direction
by the antenna gain pattern, the slant range differential across the antenna
beamwidth, and the surface scattering properties at the varying incident
angle. Although the record window was intended to be centered about the
boresight return, the 300 sec record window was actually systematically
mispositioned slightly off-center by 40 to 60 _sec. Since the gain pattern
falls sharply beyond the 3 dB beamwidth, small attitude or boresight pointing
error may cause the more pronounced gain variations in that portion of the
record window to deviate pronouncedly from the expected values. Figure C2
shows the two-way antenna gain variation caused by a location uncertainty of
5 _sec within the record window, assuming the record window is centered about
the boresight return. The uncertainties due to the variation in the antenna
gain pattern, variations in amplitude over the 20 MHz frequency band,
variations of antenna gain with temperature and flatness, and variations due
to roll angle are expected to be about + 1.5 dB.
C-4
c_
_D
" 1
Ic_
v 0c_CDC_C_LuJ
-I<ZZL_
Z
< -2
\\
\
//
/
I
60 -120
i I I . I i ' J
-00 -40 0 40 80 120 1O0
Time Delay in IPP, uSec
Figure C2. Two-Way Antenna Gain Variation
C-5
C.3 CALIBRATION CONSIDERATIONS OF THE SAR RECEIVER
Amplitude calibration requires knowledge of gain variation of the SAR receiver
and also the level of the receiver thermal noise.
C.3.1 Effect of Misposltionlng of the Record Window Upon Dynamic Range
The return signal is amplified in the SAR electronics receiver and amplitude
modulated, using sensitivity time control (STC) circuitry, to coarsely
compensate for the antenna gain pattern in range° However, the combination of
the mispositionlng of the record window and the STC greatly reduces the
dynamic range of the input signal to the data link, as illustrated in
Figure C3. For a uniform, extended target, the resulting envelope of the
return power output from the receiver is expected to vary 3 dB to 4 dB over
the swath for a centered record window (Figure C3a); and as much as 15 dB for
a mispositioned record window (Figure C3b). Therefore, if the data link
dynamic range is specified to be 20 dB, for example, then the baekscatter
coefficient of the extended target can_vary 16 dB to 17 dB and still be within
the data link dynamic range, if the record window is centered. If the record
window is not centered, as shown in Figure C3b, the backscatter coefficient
can vary only 5 dB and still fall within the data llnk dynamic range. Hence,I
the range of backscatter Coefficient which the data llnk can accommodate is
lowered by as much as ii dB to 12 dB if the record window is mispositioned as
in this example.
Even with maximum calibration efforts, the uncertainty of the combination of
the receiver gain and the STC gain (allowing for some timing offset error in
the STC waveform) is expected to be about + 0.5 dB.
C-6
-HCORRECTWINDOW F-
\/ \
/ \/ \
2 -WAY I
ANTENNA------__GAIN \
1\\, /\\
STCWAVEFORM
I
l/
\ //
_ | ' !-I 0 0 1 50
REL TIME IN IPP (BSEC)
-5
RESULTINGENVELOPE
-lO
-15
-2O
RELPWR(dB
INCORRECT
.._ WINDOW _-
/''%/ \
\\\ _ RESULTING
' \f" \\
f.-- -.,\ \ \
\ l\ I\ I\ /
v ! i
-I 50 0 1 50
REL TIME IN IPP (_SEC)
(a) Centered Record Window (b) Off-Centered Record Window
Figure C3. Reduction of Dynamic Range Caused by
Mispositioning of Record Window
C-7
C.3.2 Use of Recelve-Only Noise in Calibration
Because of a pre-launch failure, the retriggered chirp calibration pulse was
not able to be used for amplitude calibration. In the absence of this
calibration pulse, the receiver thermal noise level may be used to perform a
coarse amplitude calibration. Typically at the beginning of every SEASAT SAR
pass before the SAR transmitter was enabled, a few seconds of data were
recorded upon the high density data record (HDDR) while the 5AR electronics
receiver was enabled. This recorded, receiver noise is called the "receive-
only" noise. This receive-only noise also contains noise contributions from
the antenna, the data link, the SAR data formatter (SDF), and the HDDR
recorder. The receiver thermal noise level was measured prior to launch, so
that for the case of thermal noise dominating the receive-only noise, the use
of this noise source may permit calibration of the SAR receive gain. Since
the receiver noise is modulated by the "'V-notched" STC waveform
(Reference CI), as shown in Figure C3, the less attenuated portion (by the
STC), at the beginning of the record window contains less of other system
noises, and therefore is more reliable for receiver gain calibration.
There are several disadvntages in using the recelve-only noise in calibration:
(I) To relate power of the received signal to power of the receive-only
noise, the different spectral characteristics (as illustrated in
Figure C4) must be taken into account,
(2) Many of the system noise contributors are not as well characterized as
the receiver thermal noise,
(3) It is possible that returns from ground based radars may have interfered
with the SAR recelve-only noise to the point of complete saturation of
the data llnk (Reference C3).
C-8
SIGNALf SPECTRUM
(a
THERMALNOISE
_ .. _,SPECTRUM
Range Spectra
e_
SIGNAL SPECTRUM(SHAPED BY ANTENNA GAIN)
(b) Azimuth Spectra
Figure C4. Spectral Characteristics of Signal
and Receiver Thermal Noise
C-9
C.4 CALIBRATION CONSIDERATIONS OF THE SPACECRAFT PORTION OF THE DATA LINK
The spacecraft portion of the data link includes a modulator, an L-band to
S-band translator, an S-band transmitter, and the data link Helix antenna as
shown in Figure C5. The data link was designed to preserve the SAR return
echo, the proper phasing information, and timing signals to present to the
ground A/D converter, formatter, and recorder. The L-band return echo from
the SAR receiver is RF summed to a pseudorandom noise (PN) sequence and to a
pilot tone, translated to S-band, amplified, and radiated by the data link
(D/L) antenna. The ground station portion of the data link includes the
receiving antenna, an S-band preamplifier, multicoupler, down-converter,
multi-function receiver (MFR), and SAR demodulator. The S-band signal
transmitted from the spacecraft is received by the satellite tracking and data
acquisition network (STDN) antenna, pre-amplified at S-band, down-converted to
offset video in the SAR demodulator.
The gain of the data link was designed to be approximately unity. Actually,
the data link gain depends upon the following (Reference C2):
(i) The transmitted pilot tone level variations as referenced to the input to
the modulator/translator/transmitter,
(2) The S-band transmitter dependence upon temperature and average input
signal level, and
(3) The attenuator setting in the ii0 MHz line from the MFR to the
demodulator.
C-LO
i o -
4wc_
z
_J
-1I
I
i
tI
iI
I
I
i
Ii
I
I
I
I
t
tL _
U'h
C-II
The pilot tone level varies with temperature and the SAR receiver gain state
because an inadvertent coherent L-band leakage tone from the SAR electronics
combines with the data link pilot tone to produce a resultant pilot tone at a
different level. The automatic gain control (AGC) circuitry in the station
MFR then changes its gain correspondingly, which changes the data link gain
from unity. For the typical operational range of S/C temperatures and the
range of receiver gain settings actually used in flight, the data llnk gain is
expected to vary -0.5 dB to +2.5 dB from unity gain. An additional effect is
the small signal suppression of the pilot tone in the data link as the
strength of the return echo increases and causes saturation in the data
link. Also, the transmitter gain is temperature dependent, causing the pilot
tone level to change with the varying temperature of the S-band transmitter.
For a reasonable range of return echo strengths and observed S/C temperatures,
the gain variation is expected to be from 0. I dB to 1.8 dB. Soft saturation
of the return echo may occur in the data link, for even if the receiver gain
is correctly set, the record window misposition (shown in Figure C3) may
cause returns of sufficient strength to be compressed considerably. The
effect upon gain of the attenuator setting in the II0 MHz llne is discussed in
the next subsection.
C-12
C.5 CALIBRATIONCONSIDERATIONSOFTHEGROUNDPORTIONOF THEDATALINK
The gain variations in the ground portion of the data link must also be taken
into consideration for amplitude calibration. Figure C5 shows the ground
portion of the data link as well as the spacecraft portion. The received
signal at the ground station is known to contain aliased engineering telemetry
data which is transmitted at a higher S-band frequency, but because of the
filter characteristics of the data link and the sampling rate of the A/D
converter, the engineering telemetry spectrum is folded onto the SARecho
spectrum. The strength of the aliased signals are muchlower than the return
echo; however, for certain configurations of the spacecraft and the ground
station, the strength of the aliased signals may be on the order of, or even
higher than, the other receive-only noises over portions of the recordJ
window. This could significantly affect calibration since the receive-only
noise level is key in amplitude calibration; any variation in the noise level
propagates as amplitude calibration uncertainty. The spectrum of the aliased
engineering telemetry signal changes significantly with the telemetry mode
(real-time telemetry, ranging, and playback). The various components of the
composite engineering telemetry spectrum are represented in Figure C6a. Only
a subset of these components are ever present in any particular operational
mode. The spectrum of the data link reference signals and the receiver
thermal noise, upon which the aliased engineering telemetry spectrum is
folded, is represented in Figure C6b.
The 9m STDN antenna is intended to receive the data llnk signal when the data
link transmitter is in the high power mode; the 26m STDN antenna, for the data
link low power mode. The S/C was nominally configured in the data link high
power mode; in the low power mode, the S-band transmitter gain was attenuated
C-13
6 dB, allowing a greater dynamic range for the data link for transmitting SAR
echo data. However, in several cases, the 26mantenna was allocated to
receive the high power modesignal. For sufficiently strong returns, the
signals could approach the non-llnear portion of the preamplifier-to-down-
converter transfer.
The level of the pilot tone in the II0 MHzllne from the MFRwas intended to
be constant at the demodulator input. However_because of the difference in
the gain of individual MFR's at the stations, variations from the nominal
setting were observed during several SEASATpasses to be as muchas -3 dB to
+i dB. Also, an amplifier/attenuator combination in the 110 MHzline to
compensate for the cabling loss between the MFRand the demodulator was
observed on several occasions not to have been set correctly; this resulted in
as muchas 3 dB variation in the system gain.
C-14
°'°°'° I°" .•
_'- 16 MHz
\\k °o• o•
m
\.
CARRIER
/RANGI NG
TONES
//
,•'_•° J
• • °_ _
°.
• °
SUBCARRIER
25 kbps
fkk
k\
RIT
800 kbpsfPLAYB_CK
I
2284.5I # I J i I .... I
2287.5Frequency (MHz)
! | I
229O. 5
(a) Engineering Telemetry Spectra
/
PILOT TONE
//
PN CODE"FI- \/PRF \
6 MHz ----,-%/
FREQUENCY
MFR THERMAL NOISE
I \\
(b) Data Link Reference and Thermal Noise Spectra
Figure C6. Engineering Telemetry and Data Link Spectra
C-15
C.6 CALIBRATION CONSIDERATIONS OF THE GROUND STATION A/D CONVERTER,
DATA FORMATTER, AND DATA RECORDER
Gain variations and non-linearities are introduced at the ground station as
the data is A/D converted, formatted, and recorded. Figure C7 shows the data
flow from the output of the demodulator to the HDDR tape. The offset video
signal from the demodulator is presented to the SAR data formatter (SDF) to be
amplified, A/D converted, formatted, and recorded onto the HDDR. The SDF
front end gain state is selected to be high or low, depending upon whether the
data link power mode is high or low, respectively. It is known that the ULA
station had the incorrect SDF gain setting for the earlier passes, causing a
6 dB gain reduction, therefore, the actual SDF gain setting (hl/lo) should be
verified for every pass. Alsot measurements at the ground stations showed
that the actual SDF front end gain varied from station to station by as much
as 3 dB (Reference C3).
After the offset video signal is amplified in the front end of the SDF, it is
sampled and converted from analog to digital, using 5-bit quantlzation.
Quantization and saturation noise in the A/D converter is a function of the
relative strength of the input signal to the saturation level of the A/D
converter. Because of the gain variations in the system prior to the AD
converter, there are uncertainties in determining the signal power for which
saturation occurs. The A/D converted and formatted data is then recorded upon
the HDDR, along with other housekeeping information.
C-16
0<
p.iJ !
__<_8
z
-SI--
0(D
f_
<=>
\4
!
$ N
_ao_
5
0
_A
A
z ...1
zI.- 0z --0 _
_. I,=-
W
l,..-z _
1!
II
III
III
i
II
II
I
I
I
I
I
I
II
I
l
I
I
I
,. I
v
0
C_
,"h
b-.
..1=
N
o
<m
r_
t_
C-17
C.7 CALIBRATION CONSIDERATIONS OF THE DATA PROCESSING SUBSYSTEM
Considerations of the radiometric calibration of the SEASAT SAR system have
been confined to those elements of the system prior to, and including, the
recording of the SAR data upon the HDDR tapes. As shown in Figure CI,
portions of the data upon the HDDR tape are played back over an optical fiber
llnk to the digital processor. Further study is to be given to the effect
upon amplitude calibration caused by digital processing, such as would occur
in range and azimuth correlation.
C-18
Co8 SUMMARYOFRADIOMETRICCALILBRATIONCONSIDERATIONS
In this section, consideration has been given to those elements in the SEASAT
SARsystem, up to the HDDRrecording, which affect the amplitude calibration
of the SARdata. The amplitude calibration is affected by (I) System gain
variations, and (2) system non-linearitles, such as gain compression and
saturation. Use of the "receive-only" noise (in the absence of any other
calibration signal), as recorded at the beginning of the pass, allows one to
compensate for gain variations which occur after the SARelectronics
receiver. For example, if a system gain variation is caused by an incorrect
setting at the ground station receiver, the receive-only noise level will
reflect that samegain variation, the degree to which the gain variation may
be co=rected corresponds to the degree to which the receive-only noise level
can be characterized (Reference C4). This need to characterize the receive-
only noise requires one to pay special attention to all system noise
contributions which are sufficiently strong to appreciably alter the receive-
only noise level.
It is expected that the amplitude of the uncalibrated data will be known to
within no better than + 9 dB; however, with proper calibration procedures,
using available temperture data, S/C attitude data, pre-launch measurements,
the engineering data recorded upon the sensor data records, and configurationJ
data recorded in various logs, it is expected that the amplitude of the data
can be calibrated to within + 1.5 dB, up to the HDDR. This assumes that no
signficant non-linearities are introduced and that the signal-to-noise ratio
of the return signal is sufficiently high. Also, only SAR system controllable
error sources have been considered; sources such as atmospheric, ionospheric,
and Faraday rotation effects, and speckle have not been considered in this
report.
C-19
CI.
REFERENCES
"Cochannel Intereference Analysis BetweenSpaceborneand Terrestlal
Radars," John Nicholas, Jr., IEEE Transactions on Aerospace and
Electronics, Vol. AES-14, No. 5, September 1978.
C2. "Effect of SEASAT-A SAR Data Link Noise at Low Elevation Angles,"
B. Huneycutt, JPL IOM 3395-79-034, 16 February 1979. (JPL internal
document.)
C3. "Results of SEASAT-A SAR Station Observation at STDN Sites GDS, ULA, and
MIL," B. Huneycutt, JPL IOM 3394-78-114, 25 September 1978. (JPL
internal document.)
C4. "SEASAT-A SAR Receive-Only Noise Temperature," B. Huneycutt, JPL IOM
334.5-70, 13 April 1979. (JPL internal document.)
C--20 NA_-JI_-CGml LA Cahl
